@micdrop/elevenlabs 1.3.0 → 1.4.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +9 -8
- package/dist/index.d.mts +46 -3
- package/dist/index.d.ts +46 -3
- package/dist/index.js +99 -9
- package/dist/index.js.map +1 -1
- package/dist/index.mjs +103 -10
- package/dist/index.mjs.map +1 -1
- package/package.json +2 -2
package/README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
# @micdrop/elevenlabs
|
|
2
2
|
|
|
3
|
-
[Micdrop website](https://micdrop.dev) | [Documentation](https://micdrop.dev/docs/
|
|
3
|
+
[Micdrop website](https://micdrop.dev) | [Documentation](https://micdrop.dev/docs/text-to-speech/elevenlabs)
|
|
4
4
|
|
|
5
5
|
`@micdrop/elevenlabs` streams ElevenLabs text to speech for [@micdrop/server](https://micdrop.dev/docs/server). It works with Eleven v4 Turbo, Eleven v4, Eleven v3, Flash v2.5, Turbo v2.5 and Multilingual v2, and picks the WebSocket each model needs.
|
|
6
6
|
|
|
@@ -23,6 +23,7 @@ const tts = new ElevenLabsTTS({
|
|
|
23
23
|
voiceId: '21m00Tcm4TlvDq8ikWAM', // ElevenLabs voice ID
|
|
24
24
|
modelId: 'eleven_v4_turbo', // Optional: model to use
|
|
25
25
|
language: 'en', // Optional: language code
|
|
26
|
+
sampleRate: 24000, // Optional: 48000 by default
|
|
26
27
|
voiceSettings: {
|
|
27
28
|
stability: 0.5,
|
|
28
29
|
},
|
|
@@ -46,7 +47,7 @@ const tts = new ElevenLabsTTS({
|
|
|
46
47
|
voiceId: '21m00Tcm4TlvDq8ikWAM',
|
|
47
48
|
})
|
|
48
49
|
|
|
49
|
-
// Audio is raw PCM, 16 bits,
|
|
50
|
+
// Audio is raw PCM, 16 bits, mono, at tts.sampleRate
|
|
50
51
|
tts.on('Audio', (chunk) => console.log('Audio:', chunk.length, 'bytes'))
|
|
51
52
|
tts.on('Failed', (texts) => console.error('Failed:', texts))
|
|
52
53
|
|
|
@@ -55,12 +56,12 @@ tts.speak(Readable.from(['Hello! ', 'What can I do for you?']))
|
|
|
55
56
|
|
|
56
57
|
### Events
|
|
57
58
|
|
|
58
|
-
| Event | Payload | Description
|
|
59
|
-
| -------- | ---------- |
|
|
60
|
-
| `Audio` | `Buffer` | A chunk of audio, PCM 16 bits,
|
|
61
|
-
| `Failed` | `string[]` | The text left unspoken when `ElevenLabsTTS` gives up after its retries.
|
|
59
|
+
| Event | Payload | Description |
|
|
60
|
+
| -------- | ---------- | -------------------------------------------------------------------------------------------------------------------------- |
|
|
61
|
+
| `Audio` | `Buffer` | A chunk of audio, PCM 16 bits, mono, at `tts.sampleRate` (the `sampleRate` option, 48 kHz by default), ready to be played. |
|
|
62
|
+
| `Failed` | `string[]` | The text left unspoken when `ElevenLabsTTS` gives up after its retries. |
|
|
62
63
|
|
|
63
|
-
See the [TTS interface](https://micdrop.dev/docs/
|
|
64
|
+
See the [TTS interface](https://micdrop.dev/docs/text-to-speech/custom) for the full contract.
|
|
64
65
|
|
|
65
66
|
### Models
|
|
66
67
|
|
|
@@ -77,7 +78,7 @@ ElevenLabs counts each open connection to the Text to Dialogue WebSocket as one
|
|
|
77
78
|
|
|
78
79
|
## Documentation
|
|
79
80
|
|
|
80
|
-
Read full [documentation of the ElevenLabs integration for Micdrop](https://micdrop.dev/docs/
|
|
81
|
+
Read full [documentation of the ElevenLabs integration for Micdrop](https://micdrop.dev/docs/text-to-speech/elevenlabs) on the [website](https://micdrop.dev).
|
|
81
82
|
|
|
82
83
|
## License
|
|
83
84
|
|
package/dist/index.d.mts
CHANGED
|
@@ -1,13 +1,24 @@
|
|
|
1
|
-
import { TTS } from '@micdrop/server';
|
|
1
|
+
import { TTS, MicdropModel, MicdropVoice } from '@micdrop/server';
|
|
2
2
|
import { Readable } from 'stream';
|
|
3
3
|
import { TextToSpeechStreamRequestOutputFormat, VoiceSettings } from '@elevenlabs/elevenlabs-js/api';
|
|
4
4
|
|
|
5
5
|
type ElevenLabsModelId = 'eleven_v4_turbo' | 'eleven_v4' | 'eleven_v3_conversational' | 'eleven_v3' | 'eleven_flash_v2_5' | 'eleven_multilingual_v2' | 'eleven_turbo_v2_5';
|
|
6
|
+
/** Rates ElevenLabs streams raw PCM at. 44100 needs a Pro plan or above. */
|
|
7
|
+
type ElevenLabsSampleRate = 8000 | 16000 | 22050 | 24000 | 32000 | 44100 | 48000;
|
|
6
8
|
interface ElevenLabsTTSOptions {
|
|
7
9
|
apiKey: string;
|
|
8
10
|
voiceId: string;
|
|
9
11
|
modelId?: ElevenLabsModelId;
|
|
10
12
|
language?: string;
|
|
13
|
+
/**
|
|
14
|
+
* Rate of the audio, in Hz, 48000 by default, the highest one offered. It
|
|
15
|
+
* goes to the client as is, so the voice is never resampled on the server.
|
|
16
|
+
*/
|
|
17
|
+
sampleRate?: ElevenLabsSampleRate;
|
|
18
|
+
/**
|
|
19
|
+
* @deprecated Use `sampleRate`. The client only plays raw PCM, so only the
|
|
20
|
+
* `pcm_*` formats work.
|
|
21
|
+
*/
|
|
11
22
|
outputFormat?: TextToSpeechStreamRequestOutputFormat;
|
|
12
23
|
voiceSettings?: VoiceSettings;
|
|
13
24
|
connectionTimeout?: number;
|
|
@@ -15,7 +26,13 @@ interface ElevenLabsTTSOptions {
|
|
|
15
26
|
maxRetry?: number;
|
|
16
27
|
}
|
|
17
28
|
declare const DEFAULT_MODEL_ID = "eleven_turbo_v2_5";
|
|
18
|
-
declare const
|
|
29
|
+
declare const DEFAULT_SAMPLE_RATE: ElevenLabsSampleRate;
|
|
30
|
+
/** @deprecated Use `DEFAULT_SAMPLE_RATE` */
|
|
31
|
+
declare const DEFAULT_OUTPUT_FORMAT = "pcm_48000";
|
|
32
|
+
/** The `output_format` asked of ElevenLabs */
|
|
33
|
+
declare function getOutputFormat(options: ElevenLabsTTSOptions): TextToSpeechStreamRequestOutputFormat;
|
|
34
|
+
/** Rate of the audio ElevenLabs sends, read from its `output_format` */
|
|
35
|
+
declare function getSampleRate(options: ElevenLabsTTSOptions): number;
|
|
19
36
|
/** Eleven v3 and v4 only run on the Text to Dialogue WebSocket */
|
|
20
37
|
declare function isDialogueModel(modelId: string): boolean;
|
|
21
38
|
type ElevenLabsWebSocketMessage = ElevenLabsWebSocketAudioOutputMessage | ElevenLabsWebSocketFinalOutputMessage | ElevenLabsWebSocketErrorMessage;
|
|
@@ -68,11 +85,37 @@ type ElevenLabsDialogueMessage = {
|
|
|
68
85
|
* so the model picks the one this class drives.
|
|
69
86
|
*/
|
|
70
87
|
declare class ElevenLabsTTS extends TTS {
|
|
88
|
+
/**
|
|
89
|
+
* Lists the models that speak, with the languages each one reads.
|
|
90
|
+
*
|
|
91
|
+
* The list is fetched once and shared by every call of the process for a
|
|
92
|
+
* day. `refresh` fetches it again.
|
|
93
|
+
*/
|
|
94
|
+
static listModels(options: {
|
|
95
|
+
apiKey: string;
|
|
96
|
+
refresh?: boolean;
|
|
97
|
+
}): Promise<MicdropModel[]>;
|
|
98
|
+
/**
|
|
99
|
+
* Lists the voices the account can speak with: the premade ones, and the
|
|
100
|
+
* ones cloned, designed or added from the library.
|
|
101
|
+
*
|
|
102
|
+
* `language`, such as `fr` or `fr-FR`, keeps the voices that speak it.
|
|
103
|
+
*
|
|
104
|
+
* The list is fetched once and shared by every call of the process for a
|
|
105
|
+
* day. `refresh` fetches it again, for a voice added a moment ago.
|
|
106
|
+
*/
|
|
107
|
+
static listVoices(options: {
|
|
108
|
+
apiKey: string;
|
|
109
|
+
language?: string;
|
|
110
|
+
refresh?: boolean;
|
|
111
|
+
}): Promise<MicdropVoice[]>;
|
|
71
112
|
private readonly webSocket;
|
|
113
|
+
private readonly rate;
|
|
72
114
|
constructor(options: ElevenLabsTTSOptions);
|
|
115
|
+
get sampleRate(): number;
|
|
73
116
|
speak(textStream: Readable): void;
|
|
74
117
|
cancel(): void;
|
|
75
118
|
destroy(): void;
|
|
76
119
|
}
|
|
77
120
|
|
|
78
|
-
export { DEFAULT_MODEL_ID, DEFAULT_OUTPUT_FORMAT, type ElevenLabsDialogueClientMessage, type ElevenLabsDialogueInput, type ElevenLabsDialogueMessage, type ElevenLabsModelId, ElevenLabsTTS, type ElevenLabsTTSOptions, type ElevenLabsWebSocketAudioOutputMessage, type ElevenLabsWebSocketAudioOutputMessageAlignment, type ElevenLabsWebSocketErrorMessage, type ElevenLabsWebSocketFinalOutputMessage, type ElevenLabsWebSocketMessage, isDialogueModel };
|
|
121
|
+
export { DEFAULT_MODEL_ID, DEFAULT_OUTPUT_FORMAT, DEFAULT_SAMPLE_RATE, type ElevenLabsDialogueClientMessage, type ElevenLabsDialogueInput, type ElevenLabsDialogueMessage, type ElevenLabsModelId, type ElevenLabsSampleRate, ElevenLabsTTS, type ElevenLabsTTSOptions, type ElevenLabsWebSocketAudioOutputMessage, type ElevenLabsWebSocketAudioOutputMessageAlignment, type ElevenLabsWebSocketErrorMessage, type ElevenLabsWebSocketFinalOutputMessage, type ElevenLabsWebSocketMessage, getOutputFormat, getSampleRate, isDialogueModel };
|
package/dist/index.d.ts
CHANGED
|
@@ -1,13 +1,24 @@
|
|
|
1
|
-
import { TTS } from '@micdrop/server';
|
|
1
|
+
import { TTS, MicdropModel, MicdropVoice } from '@micdrop/server';
|
|
2
2
|
import { Readable } from 'stream';
|
|
3
3
|
import { TextToSpeechStreamRequestOutputFormat, VoiceSettings } from '@elevenlabs/elevenlabs-js/api';
|
|
4
4
|
|
|
5
5
|
type ElevenLabsModelId = 'eleven_v4_turbo' | 'eleven_v4' | 'eleven_v3_conversational' | 'eleven_v3' | 'eleven_flash_v2_5' | 'eleven_multilingual_v2' | 'eleven_turbo_v2_5';
|
|
6
|
+
/** Rates ElevenLabs streams raw PCM at. 44100 needs a Pro plan or above. */
|
|
7
|
+
type ElevenLabsSampleRate = 8000 | 16000 | 22050 | 24000 | 32000 | 44100 | 48000;
|
|
6
8
|
interface ElevenLabsTTSOptions {
|
|
7
9
|
apiKey: string;
|
|
8
10
|
voiceId: string;
|
|
9
11
|
modelId?: ElevenLabsModelId;
|
|
10
12
|
language?: string;
|
|
13
|
+
/**
|
|
14
|
+
* Rate of the audio, in Hz, 48000 by default, the highest one offered. It
|
|
15
|
+
* goes to the client as is, so the voice is never resampled on the server.
|
|
16
|
+
*/
|
|
17
|
+
sampleRate?: ElevenLabsSampleRate;
|
|
18
|
+
/**
|
|
19
|
+
* @deprecated Use `sampleRate`. The client only plays raw PCM, so only the
|
|
20
|
+
* `pcm_*` formats work.
|
|
21
|
+
*/
|
|
11
22
|
outputFormat?: TextToSpeechStreamRequestOutputFormat;
|
|
12
23
|
voiceSettings?: VoiceSettings;
|
|
13
24
|
connectionTimeout?: number;
|
|
@@ -15,7 +26,13 @@ interface ElevenLabsTTSOptions {
|
|
|
15
26
|
maxRetry?: number;
|
|
16
27
|
}
|
|
17
28
|
declare const DEFAULT_MODEL_ID = "eleven_turbo_v2_5";
|
|
18
|
-
declare const
|
|
29
|
+
declare const DEFAULT_SAMPLE_RATE: ElevenLabsSampleRate;
|
|
30
|
+
/** @deprecated Use `DEFAULT_SAMPLE_RATE` */
|
|
31
|
+
declare const DEFAULT_OUTPUT_FORMAT = "pcm_48000";
|
|
32
|
+
/** The `output_format` asked of ElevenLabs */
|
|
33
|
+
declare function getOutputFormat(options: ElevenLabsTTSOptions): TextToSpeechStreamRequestOutputFormat;
|
|
34
|
+
/** Rate of the audio ElevenLabs sends, read from its `output_format` */
|
|
35
|
+
declare function getSampleRate(options: ElevenLabsTTSOptions): number;
|
|
19
36
|
/** Eleven v3 and v4 only run on the Text to Dialogue WebSocket */
|
|
20
37
|
declare function isDialogueModel(modelId: string): boolean;
|
|
21
38
|
type ElevenLabsWebSocketMessage = ElevenLabsWebSocketAudioOutputMessage | ElevenLabsWebSocketFinalOutputMessage | ElevenLabsWebSocketErrorMessage;
|
|
@@ -68,11 +85,37 @@ type ElevenLabsDialogueMessage = {
|
|
|
68
85
|
* so the model picks the one this class drives.
|
|
69
86
|
*/
|
|
70
87
|
declare class ElevenLabsTTS extends TTS {
|
|
88
|
+
/**
|
|
89
|
+
* Lists the models that speak, with the languages each one reads.
|
|
90
|
+
*
|
|
91
|
+
* The list is fetched once and shared by every call of the process for a
|
|
92
|
+
* day. `refresh` fetches it again.
|
|
93
|
+
*/
|
|
94
|
+
static listModels(options: {
|
|
95
|
+
apiKey: string;
|
|
96
|
+
refresh?: boolean;
|
|
97
|
+
}): Promise<MicdropModel[]>;
|
|
98
|
+
/**
|
|
99
|
+
* Lists the voices the account can speak with: the premade ones, and the
|
|
100
|
+
* ones cloned, designed or added from the library.
|
|
101
|
+
*
|
|
102
|
+
* `language`, such as `fr` or `fr-FR`, keeps the voices that speak it.
|
|
103
|
+
*
|
|
104
|
+
* The list is fetched once and shared by every call of the process for a
|
|
105
|
+
* day. `refresh` fetches it again, for a voice added a moment ago.
|
|
106
|
+
*/
|
|
107
|
+
static listVoices(options: {
|
|
108
|
+
apiKey: string;
|
|
109
|
+
language?: string;
|
|
110
|
+
refresh?: boolean;
|
|
111
|
+
}): Promise<MicdropVoice[]>;
|
|
71
112
|
private readonly webSocket;
|
|
113
|
+
private readonly rate;
|
|
72
114
|
constructor(options: ElevenLabsTTSOptions);
|
|
115
|
+
get sampleRate(): number;
|
|
73
116
|
speak(textStream: Readable): void;
|
|
74
117
|
cancel(): void;
|
|
75
118
|
destroy(): void;
|
|
76
119
|
}
|
|
77
120
|
|
|
78
|
-
export { DEFAULT_MODEL_ID, DEFAULT_OUTPUT_FORMAT, type ElevenLabsDialogueClientMessage, type ElevenLabsDialogueInput, type ElevenLabsDialogueMessage, type ElevenLabsModelId, ElevenLabsTTS, type ElevenLabsTTSOptions, type ElevenLabsWebSocketAudioOutputMessage, type ElevenLabsWebSocketAudioOutputMessageAlignment, type ElevenLabsWebSocketErrorMessage, type ElevenLabsWebSocketFinalOutputMessage, type ElevenLabsWebSocketMessage, isDialogueModel };
|
|
121
|
+
export { DEFAULT_MODEL_ID, DEFAULT_OUTPUT_FORMAT, DEFAULT_SAMPLE_RATE, type ElevenLabsDialogueClientMessage, type ElevenLabsDialogueInput, type ElevenLabsDialogueMessage, type ElevenLabsModelId, type ElevenLabsSampleRate, ElevenLabsTTS, type ElevenLabsTTSOptions, type ElevenLabsWebSocketAudioOutputMessage, type ElevenLabsWebSocketAudioOutputMessageAlignment, type ElevenLabsWebSocketErrorMessage, type ElevenLabsWebSocketFinalOutputMessage, type ElevenLabsWebSocketMessage, getOutputFormat, getSampleRate, isDialogueModel };
|
package/dist/index.js
CHANGED
|
@@ -22,7 +22,10 @@ var index_exports = {};
|
|
|
22
22
|
__export(index_exports, {
|
|
23
23
|
DEFAULT_MODEL_ID: () => DEFAULT_MODEL_ID,
|
|
24
24
|
DEFAULT_OUTPUT_FORMAT: () => DEFAULT_OUTPUT_FORMAT,
|
|
25
|
+
DEFAULT_SAMPLE_RATE: () => DEFAULT_SAMPLE_RATE,
|
|
25
26
|
ElevenLabsTTS: () => ElevenLabsTTS,
|
|
27
|
+
getOutputFormat: () => getOutputFormat,
|
|
28
|
+
getSampleRate: () => getSampleRate,
|
|
26
29
|
isDialogueModel: () => isDialogueModel
|
|
27
30
|
});
|
|
28
31
|
module.exports = __toCommonJS(index_exports);
|
|
@@ -121,7 +124,16 @@ var ElevenLabsConnection = class {
|
|
|
121
124
|
|
|
122
125
|
// src/types.ts
|
|
123
126
|
var DEFAULT_MODEL_ID = "eleven_turbo_v2_5";
|
|
124
|
-
var
|
|
127
|
+
var DEFAULT_SAMPLE_RATE = 48e3;
|
|
128
|
+
var DEFAULT_OUTPUT_FORMAT = "pcm_48000";
|
|
129
|
+
function getOutputFormat(options) {
|
|
130
|
+
if (options.sampleRate) return `pcm_${options.sampleRate}`;
|
|
131
|
+
return options.outputFormat ?? `pcm_${DEFAULT_SAMPLE_RATE}`;
|
|
132
|
+
}
|
|
133
|
+
function getSampleRate(options) {
|
|
134
|
+
const match = /^pcm_(\d+)$/.exec(getOutputFormat(options));
|
|
135
|
+
return match ? Number(match[1]) : DEFAULT_SAMPLE_RATE;
|
|
136
|
+
}
|
|
125
137
|
function isDialogueModel(modelId) {
|
|
126
138
|
return modelId.startsWith("eleven_v3") || modelId.startsWith("eleven_v4");
|
|
127
139
|
}
|
|
@@ -154,10 +166,7 @@ var DialogueWebSocket = class {
|
|
|
154
166
|
this.hasSpoken = false;
|
|
155
167
|
const params = new URLSearchParams();
|
|
156
168
|
params.append("model_id", this.options.modelId ?? DEFAULT_MODEL_ID);
|
|
157
|
-
params.append(
|
|
158
|
-
"output_format",
|
|
159
|
-
this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT
|
|
160
|
-
);
|
|
169
|
+
params.append("output_format", getOutputFormat(this.options));
|
|
161
170
|
if (this.options.language) {
|
|
162
171
|
params.append("language_code", this.options.language);
|
|
163
172
|
}
|
|
@@ -373,10 +382,7 @@ var SpeechWebSocket = class {
|
|
|
373
382
|
this.heardUntil = 0;
|
|
374
383
|
const params = new URLSearchParams();
|
|
375
384
|
params.append("model_id", this.options.modelId ?? DEFAULT_MODEL_ID);
|
|
376
|
-
params.append(
|
|
377
|
-
"output_format",
|
|
378
|
-
this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT
|
|
379
|
-
);
|
|
385
|
+
params.append("output_format", getOutputFormat(this.options));
|
|
380
386
|
params.append("inactivity_timeout", WS_INACTIVITY_TIMEOUT.toString());
|
|
381
387
|
const voiceSettings = toWireVoiceSettings(this.options.voiceSettings);
|
|
382
388
|
if (voiceSettings) {
|
|
@@ -571,9 +577,56 @@ function toWireVoiceSettings(settings) {
|
|
|
571
577
|
}
|
|
572
578
|
|
|
573
579
|
// src/ElevenLabsTTS.ts
|
|
580
|
+
var CUSTOM_CATEGORIES = ["cloned", "generated", "professional"];
|
|
574
581
|
var ElevenLabsTTS = class extends import_server2.TTS {
|
|
582
|
+
/**
|
|
583
|
+
* Lists the models that speak, with the languages each one reads.
|
|
584
|
+
*
|
|
585
|
+
* The list is fetched once and shared by every call of the process for a
|
|
586
|
+
* day. `refresh` fetches it again.
|
|
587
|
+
*/
|
|
588
|
+
static listModels(options) {
|
|
589
|
+
return (0, import_server2.cachedModels)(
|
|
590
|
+
["elevenlabs"],
|
|
591
|
+
options.apiKey,
|
|
592
|
+
async () => {
|
|
593
|
+
const response = await fetch("https://api.elevenlabs.io/v1/models", {
|
|
594
|
+
headers: { "xi-api-key": options.apiKey }
|
|
595
|
+
});
|
|
596
|
+
if (!response.ok) {
|
|
597
|
+
throw new Error(`ElevenLabs answered ${response.status}`);
|
|
598
|
+
}
|
|
599
|
+
const models = await response.json();
|
|
600
|
+
return models.filter((model) => model.can_do_text_to_speech).map((model) => ({
|
|
601
|
+
id: model.model_id,
|
|
602
|
+
name: model.name,
|
|
603
|
+
description: model.description,
|
|
604
|
+
languages: model.languages?.map(({ language_id }) => language_id)
|
|
605
|
+
}));
|
|
606
|
+
},
|
|
607
|
+
options.refresh
|
|
608
|
+
);
|
|
609
|
+
}
|
|
610
|
+
/**
|
|
611
|
+
* Lists the voices the account can speak with: the premade ones, and the
|
|
612
|
+
* ones cloned, designed or added from the library.
|
|
613
|
+
*
|
|
614
|
+
* `language`, such as `fr` or `fr-FR`, keeps the voices that speak it.
|
|
615
|
+
*
|
|
616
|
+
* The list is fetched once and shared by every call of the process for a
|
|
617
|
+
* day. `refresh` fetches it again, for a voice added a moment ago.
|
|
618
|
+
*/
|
|
619
|
+
static listVoices(options) {
|
|
620
|
+
return (0, import_server2.cachedVoices)(
|
|
621
|
+
["elevenlabs"],
|
|
622
|
+
options.apiKey,
|
|
623
|
+
() => fetchVoices(options.apiKey),
|
|
624
|
+
options.refresh
|
|
625
|
+
).then((voices) => (0, import_server2.voicesForLanguage)(voices, options.language));
|
|
626
|
+
}
|
|
575
627
|
constructor(options) {
|
|
576
628
|
super();
|
|
629
|
+
this.rate = getSampleRate(options);
|
|
577
630
|
const host = {
|
|
578
631
|
audio: (chunk) => this.emit("Audio", chunk),
|
|
579
632
|
failed: (texts) => this.emit("Failed", texts),
|
|
@@ -581,6 +634,9 @@ var ElevenLabsTTS = class extends import_server2.TTS {
|
|
|
581
634
|
};
|
|
582
635
|
this.webSocket = isDialogueModel(options.modelId ?? DEFAULT_MODEL_ID) ? new DialogueWebSocket(options, host) : new SpeechWebSocket(options, host);
|
|
583
636
|
}
|
|
637
|
+
get sampleRate() {
|
|
638
|
+
return this.rate;
|
|
639
|
+
}
|
|
584
640
|
speak(textStream) {
|
|
585
641
|
this.webSocket.speak(textStream);
|
|
586
642
|
}
|
|
@@ -592,11 +648,45 @@ var ElevenLabsTTS = class extends import_server2.TTS {
|
|
|
592
648
|
super.destroy();
|
|
593
649
|
}
|
|
594
650
|
};
|
|
651
|
+
async function fetchVoices(apiKey) {
|
|
652
|
+
const voices = [];
|
|
653
|
+
let pageToken;
|
|
654
|
+
do {
|
|
655
|
+
const params = new URLSearchParams({ page_size: "100" });
|
|
656
|
+
if (pageToken) params.set("next_page_token", pageToken);
|
|
657
|
+
const response = await fetch(
|
|
658
|
+
`https://api.elevenlabs.io/v2/voices?${params}`,
|
|
659
|
+
{ headers: { "xi-api-key": apiKey } }
|
|
660
|
+
);
|
|
661
|
+
if (!response.ok) throw new Error(`ElevenLabs answered ${response.status}`);
|
|
662
|
+
const page = await response.json();
|
|
663
|
+
voices.push(...page.voices.map(toMicdropVoice));
|
|
664
|
+
pageToken = page.has_more ? page.next_page_token ?? void 0 : void 0;
|
|
665
|
+
} while (pageToken);
|
|
666
|
+
return voices;
|
|
667
|
+
}
|
|
668
|
+
function toMicdropVoice(voice) {
|
|
669
|
+
const verified = (voice.verified_languages ?? []).map(({ locale, language }) => locale || language).filter((code) => !!code);
|
|
670
|
+
const label = voice.labels?.language;
|
|
671
|
+
const languages = verified.length ? verified : label ? [label] : void 0;
|
|
672
|
+
return {
|
|
673
|
+
id: voice.voice_id,
|
|
674
|
+
name: voice.name,
|
|
675
|
+
languages: languages && [...new Set(languages)],
|
|
676
|
+
gender: (0, import_server2.voiceGender)(voice.labels?.gender),
|
|
677
|
+
description: voice.description || void 0,
|
|
678
|
+
previewUrl: voice.preview_url || void 0,
|
|
679
|
+
custom: CUSTOM_CATEGORIES.includes(voice.category ?? "")
|
|
680
|
+
};
|
|
681
|
+
}
|
|
595
682
|
// Annotate the CommonJS export names for ESM import in node:
|
|
596
683
|
0 && (module.exports = {
|
|
597
684
|
DEFAULT_MODEL_ID,
|
|
598
685
|
DEFAULT_OUTPUT_FORMAT,
|
|
686
|
+
DEFAULT_SAMPLE_RATE,
|
|
599
687
|
ElevenLabsTTS,
|
|
688
|
+
getOutputFormat,
|
|
689
|
+
getSampleRate,
|
|
600
690
|
isDialogueModel
|
|
601
691
|
});
|
|
602
692
|
//# sourceMappingURL=index.js.map
|
package/dist/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/index.ts","../src/ElevenLabsTTS.ts","../src/ElevenLabsWebSocket.ts","../src/types.ts","../src/DialogueWebSocket.ts","../src/SpeechWebSocket.ts"],"sourcesContent":["export * from './ElevenLabsTTS'\nexport * from './types'\n","import { TTS } from '@micdrop/server'\nimport { Readable } from 'stream'\nimport { DialogueWebSocket } from './DialogueWebSocket'\nimport { ElevenLabsWebSocket } from './ElevenLabsWebSocket'\nimport { SpeechWebSocket } from './SpeechWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsTTSOptions,\n isDialogueModel,\n} from './types'\n\n/**\n * ElevenLabs text to speech, streamed over the WebSocket the model runs on.\n *\n * Eleven v3 and v4 only answer on the Text to Dialogue WebSocket, the other\n * models on the Text to Speech WebSocket. The two speak different protocols,\n * so the model picks the one this class drives.\n */\nexport class ElevenLabsTTS extends TTS {\n private readonly webSocket: ElevenLabsWebSocket\n\n constructor(options: ElevenLabsTTSOptions) {\n super()\n const host = {\n audio: (chunk: Buffer) => this.emit('Audio', chunk),\n failed: (texts: string[]) => this.emit('Failed', texts),\n log: (...message: any[]) => this.log(...message),\n }\n this.webSocket = isDialogueModel(options.modelId ?? DEFAULT_MODEL_ID)\n ? new DialogueWebSocket(options, host)\n : new SpeechWebSocket(options, host)\n }\n\n speak(textStream: Readable) {\n this.webSocket.speak(textStream)\n }\n\n cancel() {\n this.webSocket.cancel()\n }\n\n destroy() {\n this.webSocket.destroy()\n super.destroy()\n }\n}\n","import { ReconnectingSocket } from '@micdrop/server'\nimport type { Readable } from 'stream'\nimport type { ElevenLabsTTSOptions } from './types'\n\nconst DEFAULT_MAX_RETRY = 3\n\n/** One of the two ElevenLabs WebSockets, driven by ElevenLabsTTS */\nexport interface ElevenLabsWebSocket {\n speak(textStream: Readable): void\n cancel(): void\n destroy(): void\n}\n\n/** What a WebSocket hands back to the ElevenLabsTTS driving it */\nexport interface ElevenLabsWebSocketHost {\n audio(chunk: Buffer): void\n failed(texts: string[]): void\n log(...message: any[]): void\n}\n\nexport interface ElevenLabsConnectionOptions {\n url: string\n tts: ElevenLabsTTSOptions\n /** First message of every connection, which sets the session up */\n init: object\n keepAlive: { interval: number; message: object }\n host: ElevenLabsWebSocketHost\n /** The connection opened, after a drop when `reconnected` is true */\n onOpen(reconnected: boolean): void\n onMessage(message: any): void\n /** Every retry failed, and the connection waits for `connect()` */\n onFailed(): void\n}\n\n/**\n * The connection of one of the two WebSockets, reconnecting on its own.\n *\n * ElevenLabs opens a connection before it checks the key and the voice, then\n * closes it with an error. Such a connection opens fine, which would start the\n * retry count of ReconnectingSocket over each time and reconnect forever. So a\n * connection closed before it proved usable counts as a failed one, and the\n * connection gives up once they exceed `maxRetry` in a row.\n */\nexport class ElevenLabsConnection {\n private socket: ReconnectingSocket\n // Whether the current connection carried audio or lived long enough for a\n // keep-alive, which means the server accepted the key and the voice\n private proven = false\n private failures = 0\n\n constructor(private readonly options: ElevenLabsConnectionOptions) {\n this.socket = this.createSocket()\n }\n\n /** Opens the connection, again after giving up, and does nothing otherwise */\n connect() {\n this.socket.connect()\n }\n\n /**\n * Resolves to false when the connection gives up or is destroyed meanwhile,\n * and right away once it gave up, until the next `connect()`: the stream\n * that was speaking then has nothing left to wait for.\n */\n ready(): Promise<boolean> {\n if (this.socket.status === 'idle') return Promise.resolve(false)\n return this.socket.ready()\n }\n\n send(message: object): boolean {\n return this.socket.send(JSON.stringify(message))\n }\n\n /** Drops the connection, with the synthesis running on it, for a new one */\n restart() {\n this.socket.restart()\n }\n\n destroy() {\n this.socket.destroy()\n }\n\n /** The current connection works, so its next drop is not a failure */\n prove() {\n this.proven = true\n this.failures = 0\n }\n\n private createSocket() {\n const { tts, keepAlive, host } = this.options\n const socket = new ReconnectingSocket({\n url: this.options.url,\n clientOptions: { headers: { 'xi-api-key': tts.apiKey } },\n connectionTimeout: tts.connectionTimeout,\n retryDelay: tts.retryDelay,\n maxRetry: tts.maxRetry,\n handshake: {\n start: (ws) => ws.send(JSON.stringify(this.options.init)),\n },\n keepAlive: {\n interval: keepAlive.interval,\n message: () => {\n this.prove()\n return JSON.stringify(keepAlive.message)\n },\n },\n log: (...message) => host.log(...message),\n })\n\n socket.on('Open', ({ reconnected }) => {\n this.proven = false\n this.options.onOpen(reconnected)\n })\n\n socket.on('Message', (data) => {\n try {\n this.options.onMessage(JSON.parse(data.toString()))\n } catch (error: any) {\n host.log('message' in error ? error.message : error)\n host.log('Event data during error:', data.toString())\n }\n })\n\n socket.on('Close', () => {\n if (this.proven) return\n this.failures++\n if (this.failures > (tts.maxRetry ?? DEFAULT_MAX_RETRY)) this.giveUp()\n })\n\n socket.on('Failed', () => this.giveUp())\n return socket\n }\n\n private giveUp() {\n this.options.host.log('Max retries reached, giving up')\n this.failures = 0\n // An idle socket replaces the one that gave up, so the next utterance\n // connects again with a full retry budget\n this.socket.destroy()\n this.socket = this.createSocket()\n this.options.onFailed()\n }\n}\n","import type {\n TextToSpeechStreamRequestOutputFormat,\n VoiceSettings,\n} from '@elevenlabs/elevenlabs-js/api'\n\nexport type ElevenLabsModelId =\n // Text to Dialogue WebSocket\n | 'eleven_v4_turbo'\n | 'eleven_v4'\n | 'eleven_v3_conversational'\n | 'eleven_v3'\n // Text to Speech WebSocket\n | 'eleven_flash_v2_5'\n | 'eleven_multilingual_v2'\n | 'eleven_turbo_v2_5'\n\nexport interface ElevenLabsTTSOptions {\n apiKey: string\n voiceId: string\n modelId?: ElevenLabsModelId\n language?: string\n outputFormat?: TextToSpeechStreamRequestOutputFormat\n voiceSettings?: VoiceSettings\n connectionTimeout?: number\n retryDelay?: number\n maxRetry?: number\n}\n\nexport const DEFAULT_MODEL_ID = 'eleven_turbo_v2_5'\nexport const DEFAULT_OUTPUT_FORMAT = 'pcm_16000'\n\n/** Eleven v3 and v4 only run on the Text to Dialogue WebSocket */\nexport function isDialogueModel(modelId: string): boolean {\n return modelId.startsWith('eleven_v3') || modelId.startsWith('eleven_v4')\n}\n\n// Text to Speech WebSocket\n\nexport type ElevenLabsWebSocketMessage =\n | ElevenLabsWebSocketAudioOutputMessage\n | ElevenLabsWebSocketFinalOutputMessage\n | ElevenLabsWebSocketErrorMessage\n\nexport interface ElevenLabsWebSocketAudioOutputMessage {\n audio: string\n normalizedAlignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n alignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n isFinal?: boolean | null\n}\n\nexport interface ElevenLabsWebSocketAudioOutputMessageAlignment {\n chars: string[]\n charStartTimesMs: number[]\n charDurationsMs: number[]\n}\n\nexport interface ElevenLabsWebSocketFinalOutputMessage {\n isFinal: boolean\n}\n\nexport interface ElevenLabsWebSocketErrorMessage {\n message: string\n error: string\n code: number\n}\n\n// Text to Dialogue WebSocket\n\nexport interface ElevenLabsDialogueClientMessage {\n voices?: string[]\n voice_settings?: { stability?: number }\n inputs?: ElevenLabsDialogueInput[]\n flush?: boolean\n keep_alive?: boolean\n}\n\nexport interface ElevenLabsDialogueInput {\n text: string\n voice_id: string\n new_turn?: boolean\n}\n\nexport type ElevenLabsDialogueMessage =\n | { audio: string }\n | { is_final_audio_for_turn: boolean }\n | { is_final: boolean }\n | ElevenLabsWebSocketErrorMessage\n","import { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n DEFAULT_OUTPUT_FORMAT,\n ElevenLabsDialogueClientMessage,\n ElevenLabsDialogueMessage,\n ElevenLabsTTSOptions,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-dialogue/ttd-websocket\n\n// The server closes a connection that hears nothing for 20 seconds\nconst KEEP_ALIVE_INTERVAL = 15000\n\n// Punctuation closing a sentence, with the space that follows it. An ellipsis\n// is a pause within the sentence, which v4 performs, so it does not close one.\nconst SENTENCE_END = /(?:[!?]+|(?<!\\.)\\.(?!\\.))\\s+/g\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `streaming`: the text stream is still coming\n * - `draining`: the text stream ended, the audio of its flushes is coming\n */\ntype UtteranceState = 'idle' | 'streaming' | 'draining'\n\n/**\n * The Text to Dialogue WebSocket, the only one serving Eleven v3 and v4.\n *\n * The server holds text back until it has about 40 characters and 8 words,\n * which delays a short first sentence such as \"Hi there!\". Flushing at the end\n * of every sentence starts its audio right away.\n *\n * Nothing in the protocol stops a synthesis once it is sent, so cancelling\n * drops the connection and opens a fresh one. It opens in under 100 ms, while\n * the user is still speaking.\n */\nexport class DialogueWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private unflushed = false // Whether text was sent since the last flush\n private pendingFlushes = 0 // Flushes whose audio is not complete yet\n private newTurn = false // Whether the next input starts a new utterance\n private hasSpoken = false // Whether the connection received any text\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append(\n 'output_format',\n this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT\n )\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n // Register the voice. Dialogue models only read the stability.\n const stability = this.options.voiceSettings?.stability\n const init: ElevenLabsDialogueClientMessage = {\n voices: [this.options.voiceId],\n voice_settings: stability === undefined ? undefined : { stability },\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-dialogue/stream-input?${params.toString()}`,\n tts: options,\n init,\n keepAlive: {\n interval: KEEP_ALIVE_INTERVAL,\n message: { keep_alive: true },\n },\n host,\n onOpen: () => this.onOpen(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say, for the same\n // reason as on the Text to Speech WebSocket: a stream that never carries a\n // word must not cut off the sentence still being spoken.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'streaming'\n this.newTurn = true\n return true\n }\n\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendText(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.endText()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n if (this.textBuffer.trim()) {\n this.sendText(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.endText()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n private onOpen() {\n this.hasSpoken = false\n if (this.state === 'idle') return\n // The flushes in flight were lost with the previous connection, so the\n // utterance is spoken again from its start\n const text = this.textSent\n this.textSent = ''\n this.unflushed = false\n this.pendingFlushes = 0\n if (text.trim()) {\n this.log('Sending text again')\n this.sendText(text)\n }\n if (this.state === 'draining') this.flush()\n this.checkEnded()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle' ? '' : this.textSent + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.unflushed = false\n this.pendingFlushes = 0\n }\n\n /** The text stream is over: flushes what the server holds */\n private endText() {\n this.state = 'draining'\n this.flush()\n this.checkEnded()\n }\n\n private onMessage(message: ElevenLabsDialogueMessage) {\n if ('audio' in message && message.audio) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n } else {\n this.log(`Received audio chunk (${chunk.length} bytes)`)\n this.host.audio(chunk)\n }\n }\n if (\n 'is_final_audio_for_turn' in message &&\n message.is_final_audio_for_turn\n ) {\n this.pendingFlushes = Math.max(0, this.pendingFlushes - 1)\n this.checkEnded()\n }\n // Sent before the server closes the connection, for a wrong key, an\n // unknown voice or no dialogue session left\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n /** Sends text, flushing after each sentence it completes */\n private sendText(text: string) {\n let start = 0\n for (const match of text.matchAll(SENTENCE_END)) {\n const end = match.index! + match[0].length\n this.sendInput(text.slice(start, end), true)\n start = end\n }\n if (start < text.length) {\n this.sendInput(text.slice(start), false)\n }\n }\n\n private sendInput(text: string, flush: boolean) {\n this.textSent += text\n if (!text.trim()) return\n this.send({\n inputs: [\n {\n text,\n voice_id: this.options.voiceId,\n // Resets the prosody between two utterances\n new_turn: (this.newTurn && this.hasSpoken) || undefined,\n },\n ],\n flush: flush || undefined,\n })\n this.log(`Sent text${flush ? ' and flushed' : ''}: \"${text}\"`)\n this.newTurn = false\n this.hasSpoken = true\n if (flush) {\n this.pendingFlushes++\n this.unflushed = false\n } else {\n this.unflushed = true\n }\n }\n\n /** Generates the text held by the server, if any */\n private flush() {\n if (!this.unflushed) return\n this.send({ flush: true })\n this.pendingFlushes++\n this.unflushed = false\n this.log('Flushed text')\n }\n\n /** Ends the utterance once its text is over and all its audio came */\n private checkEnded() {\n if (this.state !== 'draining') return\n if (this.unflushed || this.pendingFlushes > 0) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n // Lost when the connection just dropped, which is fine: the reconnection\n // sends the utterance again from textSent\n private send(message: ElevenLabsDialogueClientMessage) {\n this.connection.send(message)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n","import type { VoiceSettings } from '@elevenlabs/elevenlabs-js/api'\nimport { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n DEFAULT_OUTPUT_FORMAT,\n ElevenLabsTTSOptions,\n ElevenLabsWebSocketAudioOutputMessage,\n ElevenLabsWebSocketMessage,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-speech/v-1-text-to-speech-voice-id-stream-input\n\nconst WS_INACTIVITY_TIMEOUT = 180\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `speaking`: the text stream is still coming\n * - `flushed`: the text stream ended and was flushed, its audio is coming\n */\ntype UtteranceState = 'idle' | 'speaking' | 'flushed'\n\n/**\n * The Text to Speech WebSocket, for Flash, Multilingual v2 and Turbo v2.5.\n *\n * Its messages carry no utterance ID, and nothing in the protocol stops a\n * synthesis once it is sent. So cancelling drops the connection and opens a\n * fresh one, while the user is still speaking, and the audio of the cancelled\n * utterance is never heard of again.\n */\nexport class SpeechWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private heardUntil = 0 // Index in textSent where the audio received ends\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n // Build query params\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append(\n 'output_format',\n this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT\n )\n params.append('inactivity_timeout', WS_INACTIVITY_TIMEOUT.toString())\n const voiceSettings = toWireVoiceSettings(this.options.voiceSettings)\n if (voiceSettings) {\n params.append('voice_settings', JSON.stringify(voiceSettings))\n }\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-speech/${this.options.voiceId}/stream-input?${params.toString()}`,\n tts: options,\n // Initialization message with voice settings\n init: { text: ' ', voice_settings: voiceSettings },\n keepAlive: {\n interval: (WS_INACTIVITY_TIMEOUT - 1) * 1000,\n message: { text: ' ' },\n },\n host,\n onOpen: () => this.resume(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say.\n //\n // These fields track the utterance being spoken right now, and the\n // comparison between what was sent and what came back as audio is what\n // says where it is. Resetting them for a stream that never carries a word,\n // which is what an answer skipped by a tool or by onBeforeAnswer hands\n // over, loses the state of a sentence still being spoken and flushes it\n // for nothing.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'speaking'\n return true\n }\n\n // Forward text chunks coming from the caller to ElevenLabs\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendTranscript(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.flush()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n // Send last buffered text\n if (this.textBuffer.trim()) {\n this.sendTranscript(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.flush()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n /** Flushes buffered text and marks the end of the utterance */\n private flush() {\n this.state = 'flushed'\n this.connection.send({ text: ' ', flush: true })\n this.log('Flushed text')\n this.checkEnded()\n }\n\n /**\n * Sends the text of the utterance in flight again on a new connection. The\n * server lost what it was generating, so it goes on from the audio received.\n */\n private resume() {\n if (this.state === 'idle') return\n // Audio can stop inside a word, which goes again whole rather than from\n // its middle\n const wordStart = this.textSent.lastIndexOf(' ', this.heardUntil - 1) + 1\n const unheard = this.textSent.slice(wordStart)\n this.textSent = ''\n this.heardUntil = 0\n if (unheard.trim()) {\n this.log('Sending text chunks again')\n this.sendTranscript(unheard)\n }\n if (this.state === 'flushed') this.flush()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle'\n ? ''\n : this.textSent.slice(this.heardUntil) + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.heardUntil = 0\n }\n\n private onMessage(message: ElevenLabsWebSocketMessage) {\n if ('audio' in message && message.audio) {\n this.processAudioMessage(message)\n }\n if ('isFinal' in message && message.isFinal && this.state !== 'idle') {\n this.log('Audio ended')\n this.resetUtterance()\n }\n // Sent before the server closes the connection, for a wrong key or an\n // unknown voice\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n private processAudioMessage(message: ElevenLabsWebSocketAudioOutputMessage) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n return\n }\n\n // Check text of received audio chunk\n if (message.alignment) {\n const chunkText = message.alignment.chars.join('')\n\n // A connection carries one utterance at a time, since a cancel or a\n // new utterance opens another, so its audio is always wanted. The\n // alignment only tells how far the voice got, for a reconnection. It\n // follows the text sent, a few characters aside when the voice skips\n // or rewrites some, and is placed where it fits next.\n const end = findHeardEnd(this.textSent, this.heardUntil, chunkText)\n if (end === -1) {\n this.log(`Audio chunk off the text sent (\"${chunkText}\")`)\n } else {\n this.log(`Received audio chunk (\"${chunkText}\")`)\n this.heardUntil = end\n }\n }\n\n // Send to audio stream\n this.host.audio(chunk)\n\n // ElevenLabs doesn't send isFinal after a flush, so the end of the\n // utterance is when its audio covers all its text\n this.checkEnded()\n }\n\n private checkEnded() {\n if (this.state !== 'flushed') return\n if (this.textSent.slice(this.heardUntil).trim()) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n private sendTranscript(text: string) {\n // Counted as sent even when the connection just dropped, so the\n // reconnection sends it again\n this.textSent += text\n if (!this.connection.send({ text, try_trigger_generation: true })) return\n this.log(`Sent transcript: \"${text}\"`)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n\n/**\n * Where `heard` ends in `text`, at the first place from `from` where the text\n * goes on with it, or -1 when it does nowhere\n */\nfunction findHeardEnd(text: string, from: number, heard: string): number {\n for (let start = from; start < text.length; start++) {\n const end = endOfPrefix(text, start, heard)\n if (end !== -1) return end\n }\n return -1\n}\n\n/**\n * Where `heard` ends in `text`, read from `from` and compared without\n * whitespace, or -1 when the text does not go on with it there.\n */\nfunction endOfPrefix(text: string, from: number, heard: string): number {\n let index = from\n for (let i = 0; i < heard.length; i++) {\n if (isSpace(heard[i])) continue\n while (index < text.length && isSpace(text[index])) index++\n if (text[index] !== heard[i]) return -1\n index++\n }\n return index\n}\n\nfunction isSpace(char: string) {\n return char === ' ' || char === '\\t' || char === '\\n' || char === '\\r'\n}\n\n/**\n * Voice settings as the WebSocket reads them.\n *\n * The SDK types them in camel case and the WebSocket reads snake case, so\n * similarityBoost and useSpeakerBoost would be ignored if sent as they are.\n */\nfunction toWireVoiceSettings(settings?: VoiceSettings) {\n if (!settings) return undefined\n const { similarityBoost, useSpeakerBoost, ...rest } = settings\n return {\n similarity_boost: similarityBoost,\n use_speaker_boost: useSpeakerBoost,\n ...rest,\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACAA,IAAAA,iBAAoB;;;ACApB,oBAAmC;AAInC,IAAM,oBAAoB;AAuCnB,IAAM,uBAAN,MAA2B;AAAA,EAOhC,YAA6B,SAAsC;AAAtC;AAH7B;AAAA;AAAA,SAAQ,SAAS;AACjB,SAAQ,WAAW;AAGjB,SAAK,SAAS,KAAK,aAAa;AAAA,EAClC;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,QAA0B;AACxB,QAAI,KAAK,OAAO,WAAW,OAAQ,QAAO,QAAQ,QAAQ,KAAK;AAC/D,WAAO,KAAK,OAAO,MAAM;AAAA,EAC3B;AAAA,EAEA,KAAK,SAA0B;AAC7B,WAAO,KAAK,OAAO,KAAK,KAAK,UAAU,OAAO,CAAC;AAAA,EACjD;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA,EAEA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,SAAS;AACd,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,eAAe;AACrB,UAAM,EAAE,KAAK,WAAW,KAAK,IAAI,KAAK;AACtC,UAAM,SAAS,IAAI,iCAAmB;AAAA,MACpC,KAAK,KAAK,QAAQ;AAAA,MAClB,eAAe,EAAE,SAAS,EAAE,cAAc,IAAI,OAAO,EAAE;AAAA,MACvD,mBAAmB,IAAI;AAAA,MACvB,YAAY,IAAI;AAAA,MAChB,UAAU,IAAI;AAAA,MACd,WAAW;AAAA,QACT,OAAO,CAAC,OAAO,GAAG,KAAK,KAAK,UAAU,KAAK,QAAQ,IAAI,CAAC;AAAA,MAC1D;AAAA,MACA,WAAW;AAAA,QACT,UAAU,UAAU;AAAA,QACpB,SAAS,MAAM;AACb,eAAK,MAAM;AACX,iBAAO,KAAK,UAAU,UAAU,OAAO;AAAA,QACzC;AAAA,MACF;AAAA,MACA,KAAK,IAAI,YAAY,KAAK,IAAI,GAAG,OAAO;AAAA,IAC1C,CAAC;AAED,WAAO,GAAG,QAAQ,CAAC,EAAE,YAAY,MAAM;AACrC,WAAK,SAAS;AACd,WAAK,QAAQ,OAAO,WAAW;AAAA,IACjC,CAAC;AAED,WAAO,GAAG,WAAW,CAAC,SAAS;AAC7B,UAAI;AACF,aAAK,QAAQ,UAAU,KAAK,MAAM,KAAK,SAAS,CAAC,CAAC;AAAA,MACpD,SAAS,OAAY;AACnB,aAAK,IAAI,aAAa,QAAQ,MAAM,UAAU,KAAK;AACnD,aAAK,IAAI,4BAA4B,KAAK,SAAS,CAAC;AAAA,MACtD;AAAA,IACF,CAAC;AAED,WAAO,GAAG,SAAS,MAAM;AACvB,UAAI,KAAK,OAAQ;AACjB,WAAK;AACL,UAAI,KAAK,YAAY,IAAI,YAAY,mBAAoB,MAAK,OAAO;AAAA,IACvE,CAAC;AAED,WAAO,GAAG,UAAU,MAAM,KAAK,OAAO,CAAC;AACvC,WAAO;AAAA,EACT;AAAA,EAEQ,SAAS;AACf,SAAK,QAAQ,KAAK,IAAI,gCAAgC;AACtD,SAAK,WAAW;AAGhB,SAAK,OAAO,QAAQ;AACpB,SAAK,SAAS,KAAK,aAAa;AAChC,SAAK,QAAQ,SAAS;AAAA,EACxB;AACF;;;AClHO,IAAM,mBAAmB;AACzB,IAAM,wBAAwB;AAG9B,SAAS,gBAAgB,SAA0B;AACxD,SAAO,QAAQ,WAAW,WAAW,KAAK,QAAQ,WAAW,WAAW;AAC1E;;;ACjBA,IAAM,sBAAsB;AAI5B,IAAM,eAAe;AAsBd,IAAM,oBAAN,MAAuD;AAAA;AAAA,EAgB5D,YACmB,SACA,MACjB;AAFiB;AACA;AAhBnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,YAAY;AACpB;AAAA,SAAQ,iBAAiB;AACzB;AAAA,SAAQ,UAAU;AAClB;AAAA,SAAQ,YAAY;AAMlB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO;AAAA,MACL;AAAA,MACA,KAAK,QAAQ,gBAAgB;AAAA,IAC/B;AACA,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAGA,UAAM,YAAY,KAAK,QAAQ,eAAe;AAC9C,UAAM,OAAwC;AAAA,MAC5C,QAAQ,CAAC,KAAK,QAAQ,OAAO;AAAA,MAC7B,gBAAgB,cAAc,SAAY,SAAY,EAAE,UAAU;AAAA,IACpE;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,4DAA4D,OAAO,SAAS,CAAC;AAAA,MAClF,KAAK;AAAA,MACL;AAAA,MACA,WAAW;AAAA,QACT,UAAU;AAAA,QACV,SAAS,EAAE,YAAY,KAAK;AAAA,MAC9B;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAKd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,WAAK,UAAU;AACf,aAAO;AAAA,IACT;AAEA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,SAAS,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AACtD,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAClC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,SAAS,KAAK,aAAa,GAAG;AAAA,MACrC;AACA,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEQ,SAAS;AACf,SAAK,YAAY;AACjB,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,OAAO,KAAK;AAClB,SAAK,WAAW;AAChB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AACtB,QAAI,KAAK,KAAK,GAAG;AACf,WAAK,IAAI,oBAAoB;AAC7B,WAAK,SAAS,IAAI;AAAA,IACpB;AACA,QAAI,KAAK,UAAU,WAAY,MAAK,MAAM;AAC1C,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SAAS,KAAK,KAAK,WAAW,KAAK;AACpD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AAAA,EACxB;AAAA;AAAA,EAGQ,UAAU;AAChB,SAAK,QAAQ;AACb,SAAK,MAAM;AACX,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,UAAU,SAAoC;AACpD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,YAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,WAAK,WAAW,MAAM;AACtB,UAAI,KAAK,UAAU,QAAQ;AACzB,aAAK,IAAI,6CAA6C;AAAA,MACxD,OAAO;AACL,aAAK,IAAI,yBAAyB,MAAM,MAAM,SAAS;AACvD,aAAK,KAAK,MAAM,KAAK;AAAA,MACvB;AAAA,IACF;AACA,QACE,6BAA6B,WAC7B,QAAQ,yBACR;AACA,WAAK,iBAAiB,KAAK,IAAI,GAAG,KAAK,iBAAiB,CAAC;AACzD,WAAK,WAAW;AAAA,IAClB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA;AAAA,EAGQ,SAAS,MAAc;AAC7B,QAAI,QAAQ;AACZ,eAAW,SAAS,KAAK,SAAS,YAAY,GAAG;AAC/C,YAAM,MAAM,MAAM,QAAS,MAAM,CAAC,EAAE;AACpC,WAAK,UAAU,KAAK,MAAM,OAAO,GAAG,GAAG,IAAI;AAC3C,cAAQ;AAAA,IACV;AACA,QAAI,QAAQ,KAAK,QAAQ;AACvB,WAAK,UAAU,KAAK,MAAM,KAAK,GAAG,KAAK;AAAA,IACzC;AAAA,EACF;AAAA,EAEQ,UAAU,MAAc,OAAgB;AAC9C,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,KAAK,EAAG;AAClB,SAAK,KAAK;AAAA,MACR,QAAQ;AAAA,QACN;AAAA,UACE;AAAA,UACA,UAAU,KAAK,QAAQ;AAAA;AAAA,UAEvB,UAAW,KAAK,WAAW,KAAK,aAAc;AAAA,QAChD;AAAA,MACF;AAAA,MACA,OAAO,SAAS;AAAA,IAClB,CAAC;AACD,SAAK,IAAI,YAAY,QAAQ,iBAAiB,EAAE,MAAM,IAAI,GAAG;AAC7D,SAAK,UAAU;AACf,SAAK,YAAY;AACjB,QAAI,OAAO;AACT,WAAK;AACL,WAAK,YAAY;AAAA,IACnB,OAAO;AACL,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGQ,QAAQ;AACd,QAAI,CAAC,KAAK,UAAW;AACrB,SAAK,KAAK,EAAE,OAAO,KAAK,CAAC;AACzB,SAAK;AACL,SAAK,YAAY;AACjB,SAAK,IAAI,cAAc;AAAA,EACzB;AAAA;AAAA,EAGQ,aAAa;AACnB,QAAI,KAAK,UAAU,WAAY;AAC/B,QAAI,KAAK,aAAa,KAAK,iBAAiB,EAAG;AAC/C,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA;AAAA,EAIQ,KAAK,SAA0C;AACrD,SAAK,WAAW,KAAK,OAAO;AAAA,EAC9B;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;;;AC3RA,IAAM,wBAAwB;AAmBvB,IAAM,kBAAN,MAAqD;AAAA;AAAA,EAa1D,YACmB,SACA,MACjB;AAFiB;AACA;AAbnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,aAAa;AAOnB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO;AAAA,MACL;AAAA,MACA,KAAK,QAAQ,gBAAgB;AAAA,IAC/B;AACA,WAAO,OAAO,sBAAsB,sBAAsB,SAAS,CAAC;AACpE,UAAM,gBAAgB,oBAAoB,KAAK,QAAQ,aAAa;AACpE,QAAI,eAAe;AACjB,aAAO,OAAO,kBAAkB,KAAK,UAAU,aAAa,CAAC;AAAA,IAC/D;AACA,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,6CAA6C,KAAK,QAAQ,OAAO,iBAAiB,OAAO,SAAS,CAAC;AAAA,MACxG,KAAK;AAAA;AAAA,MAEL,MAAM,EAAE,MAAM,KAAK,gBAAgB,cAAc;AAAA,MACjD,WAAW;AAAA,QACT,WAAW,wBAAwB,KAAK;AAAA,QACxC,SAAS,EAAE,MAAM,IAAI;AAAA,MACvB;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAUd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,aAAO;AAAA,IACT;AAGA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,eAAe,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AAC5D,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAElC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,eAAe,KAAK,aAAa,GAAG;AAAA,MAC3C;AACA,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA;AAAA,EAGQ,QAAQ;AACd,SAAK,QAAQ;AACb,SAAK,WAAW,KAAK,EAAE,MAAM,KAAK,OAAO,KAAK,CAAC;AAC/C,SAAK,IAAI,cAAc;AACvB,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMQ,SAAS;AACf,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,YAAY,KAAK,SAAS,YAAY,KAAK,KAAK,aAAa,CAAC,IAAI;AACxE,UAAM,UAAU,KAAK,SAAS,MAAM,SAAS;AAC7C,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,QAAI,QAAQ,KAAK,GAAG;AAClB,WAAK,IAAI,2BAA2B;AACpC,WAAK,eAAe,OAAO;AAAA,IAC7B;AACA,QAAI,KAAK,UAAU,UAAW,MAAK,MAAM;AAAA,EAC3C;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SACX,KACA,KAAK,SAAS,MAAM,KAAK,UAAU,IAAI,KAAK;AAClD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,aAAa;AAAA,EACpB;AAAA,EAEQ,UAAU,SAAqC;AACrD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,WAAK,oBAAoB,OAAO;AAAA,IAClC;AACA,QAAI,aAAa,WAAW,QAAQ,WAAW,KAAK,UAAU,QAAQ;AACpE,WAAK,IAAI,aAAa;AACtB,WAAK,eAAe;AAAA,IACtB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA,EAEQ,oBAAoB,SAAgD;AAC1E,UAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,SAAK,WAAW,MAAM;AACtB,QAAI,KAAK,UAAU,QAAQ;AACzB,WAAK,IAAI,6CAA6C;AACtD;AAAA,IACF;AAGA,QAAI,QAAQ,WAAW;AACrB,YAAM,YAAY,QAAQ,UAAU,MAAM,KAAK,EAAE;AAOjD,YAAM,MAAM,aAAa,KAAK,UAAU,KAAK,YAAY,SAAS;AAClE,UAAI,QAAQ,IAAI;AACd,aAAK,IAAI,mCAAmC,SAAS,IAAI;AAAA,MAC3D,OAAO;AACL,aAAK,IAAI,0BAA0B,SAAS,IAAI;AAChD,aAAK,aAAa;AAAA,MACpB;AAAA,IACF;AAGA,SAAK,KAAK,MAAM,KAAK;AAIrB,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,aAAa;AACnB,QAAI,KAAK,UAAU,UAAW;AAC9B,QAAI,KAAK,SAAS,MAAM,KAAK,UAAU,EAAE,KAAK,EAAG;AACjD,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA,EAEQ,eAAe,MAAc;AAGnC,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,WAAW,KAAK,EAAE,MAAM,wBAAwB,KAAK,CAAC,EAAG;AACnE,SAAK,IAAI,qBAAqB,IAAI,GAAG;AAAA,EACvC;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;AAMA,SAAS,aAAa,MAAc,MAAc,OAAuB;AACvE,WAAS,QAAQ,MAAM,QAAQ,KAAK,QAAQ,SAAS;AACnD,UAAM,MAAM,YAAY,MAAM,OAAO,KAAK;AAC1C,QAAI,QAAQ,GAAI,QAAO;AAAA,EACzB;AACA,SAAO;AACT;AAMA,SAAS,YAAY,MAAc,MAAc,OAAuB;AACtE,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,QAAI,QAAQ,MAAM,CAAC,CAAC,EAAG;AACvB,WAAO,QAAQ,KAAK,UAAU,QAAQ,KAAK,KAAK,CAAC,EAAG;AACpD,QAAI,KAAK,KAAK,MAAM,MAAM,CAAC,EAAG,QAAO;AACrC;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,QAAQ,MAAc;AAC7B,SAAO,SAAS,OAAO,SAAS,OAAQ,SAAS,QAAQ,SAAS;AACpE;AAQA,SAAS,oBAAoB,UAA0B;AACrD,MAAI,CAAC,SAAU,QAAO;AACtB,QAAM,EAAE,iBAAiB,iBAAiB,GAAG,KAAK,IAAI;AACtD,SAAO;AAAA,IACL,kBAAkB;AAAA,IAClB,mBAAmB;AAAA,IACnB,GAAG;AAAA,EACL;AACF;;;AJnTO,IAAM,gBAAN,cAA4B,mBAAI;AAAA,EAGrC,YAAY,SAA+B;AACzC,UAAM;AACN,UAAM,OAAO;AAAA,MACX,OAAO,CAAC,UAAkB,KAAK,KAAK,SAAS,KAAK;AAAA,MAClD,QAAQ,CAAC,UAAoB,KAAK,KAAK,UAAU,KAAK;AAAA,MACtD,KAAK,IAAI,YAAmB,KAAK,IAAI,GAAG,OAAO;AAAA,IACjD;AACA,SAAK,YAAY,gBAAgB,QAAQ,WAAW,gBAAgB,IAChE,IAAI,kBAAkB,SAAS,IAAI,IACnC,IAAI,gBAAgB,SAAS,IAAI;AAAA,EACvC;AAAA,EAEA,MAAM,YAAsB;AAC1B,SAAK,UAAU,MAAM,UAAU;AAAA,EACjC;AAAA,EAEA,SAAS;AACP,SAAK,UAAU,OAAO;AAAA,EACxB;AAAA,EAEA,UAAU;AACR,SAAK,UAAU,QAAQ;AACvB,UAAM,QAAQ;AAAA,EAChB;AACF;","names":["import_server"]}
|
|
1
|
+
{"version":3,"sources":["../src/index.ts","../src/ElevenLabsTTS.ts","../src/ElevenLabsWebSocket.ts","../src/types.ts","../src/DialogueWebSocket.ts","../src/SpeechWebSocket.ts"],"sourcesContent":["export * from './ElevenLabsTTS'\nexport * from './types'\n","import {\n cachedModels,\n cachedVoices,\n MicdropModel,\n MicdropVoice,\n TTS,\n voiceGender,\n voicesForLanguage,\n} from '@micdrop/server'\nimport { Readable } from 'stream'\nimport { DialogueWebSocket } from './DialogueWebSocket'\nimport { ElevenLabsWebSocket } from './ElevenLabsWebSocket'\nimport { SpeechWebSocket } from './SpeechWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsTTSOptions,\n getSampleRate,\n isDialogueModel,\n} from './types'\n\n/** A voice as the ElevenLabs API lists it, the fields read here */\ninterface ElevenLabsVoice {\n voice_id: string\n name: string\n category?: string\n description?: string | null\n preview_url?: string | null\n labels?: Record<string, string | undefined>\n verified_languages?: Array<{ language?: string; locale?: string | null }>\n}\n\n// Voices made in the account rather than offered to everyone\nconst CUSTOM_CATEGORIES = ['cloned', 'generated', 'professional']\n\n/**\n * ElevenLabs text to speech, streamed over the WebSocket the model runs on.\n *\n * Eleven v3 and v4 only answer on the Text to Dialogue WebSocket, the other\n * models on the Text to Speech WebSocket. The two speak different protocols,\n * so the model picks the one this class drives.\n */\nexport class ElevenLabsTTS extends TTS {\n /**\n * Lists the models that speak, with the languages each one reads.\n *\n * The list is fetched once and shared by every call of the process for a\n * day. `refresh` fetches it again.\n */\n static listModels(options: {\n apiKey: string\n refresh?: boolean\n }): Promise<MicdropModel[]> {\n return cachedModels(\n ['elevenlabs'],\n options.apiKey,\n async () => {\n const response = await fetch('https://api.elevenlabs.io/v1/models', {\n headers: { 'xi-api-key': options.apiKey },\n })\n if (!response.ok) {\n throw new Error(`ElevenLabs answered ${response.status}`)\n }\n const models = (await response.json()) as Array<{\n model_id: string\n name?: string\n description?: string\n can_do_text_to_speech?: boolean\n languages?: Array<{ language_id: string }>\n }>\n return models\n .filter((model) => model.can_do_text_to_speech)\n .map((model) => ({\n id: model.model_id,\n name: model.name,\n description: model.description,\n languages: model.languages?.map(({ language_id }) => language_id),\n }))\n },\n options.refresh\n )\n }\n\n /**\n * Lists the voices the account can speak with: the premade ones, and the\n * ones cloned, designed or added from the library.\n *\n * `language`, such as `fr` or `fr-FR`, keeps the voices that speak it.\n *\n * The list is fetched once and shared by every call of the process for a\n * day. `refresh` fetches it again, for a voice added a moment ago.\n */\n static listVoices(options: {\n apiKey: string\n language?: string\n refresh?: boolean\n }): Promise<MicdropVoice[]> {\n return cachedVoices(\n ['elevenlabs'],\n options.apiKey,\n () => fetchVoices(options.apiKey),\n options.refresh\n ).then((voices) => voicesForLanguage(voices, options.language))\n }\n\n private readonly webSocket: ElevenLabsWebSocket\n private readonly rate: number\n\n constructor(options: ElevenLabsTTSOptions) {\n super()\n this.rate = getSampleRate(options)\n const host = {\n audio: (chunk: Buffer) => this.emit('Audio', chunk),\n failed: (texts: string[]) => this.emit('Failed', texts),\n log: (...message: any[]) => this.log(...message),\n }\n this.webSocket = isDialogueModel(options.modelId ?? DEFAULT_MODEL_ID)\n ? new DialogueWebSocket(options, host)\n : new SpeechWebSocket(options, host)\n }\n\n get sampleRate() {\n return this.rate\n }\n\n speak(textStream: Readable) {\n this.webSocket.speak(textStream)\n }\n\n cancel() {\n this.webSocket.cancel()\n }\n\n destroy() {\n this.webSocket.destroy()\n super.destroy()\n }\n}\n\n/** Reads every page of the voices of the account */\nasync function fetchVoices(apiKey: string): Promise<MicdropVoice[]> {\n const voices: MicdropVoice[] = []\n let pageToken: string | undefined\n do {\n const params = new URLSearchParams({ page_size: '100' })\n if (pageToken) params.set('next_page_token', pageToken)\n const response = await fetch(\n `https://api.elevenlabs.io/v2/voices?${params}`,\n { headers: { 'xi-api-key': apiKey } }\n )\n if (!response.ok) throw new Error(`ElevenLabs answered ${response.status}`)\n const page = (await response.json()) as {\n voices: ElevenLabsVoice[]\n has_more?: boolean\n next_page_token?: string | null\n }\n voices.push(...page.voices.map(toMicdropVoice))\n pageToken = page.has_more ? (page.next_page_token ?? undefined) : undefined\n } while (pageToken)\n return voices\n}\n\nfunction toMicdropVoice(voice: ElevenLabsVoice): MicdropVoice {\n const verified = (voice.verified_languages ?? [])\n .map(({ locale, language }) => locale || language)\n .filter((code): code is string => !!code)\n const label = voice.labels?.language\n const languages = verified.length ? verified : label ? [label] : undefined\n return {\n id: voice.voice_id,\n name: voice.name,\n languages: languages && [...new Set(languages)],\n gender: voiceGender(voice.labels?.gender),\n description: voice.description || undefined,\n previewUrl: voice.preview_url || undefined,\n custom: CUSTOM_CATEGORIES.includes(voice.category ?? ''),\n }\n}\n","import { ReconnectingSocket } from '@micdrop/server'\nimport type { Readable } from 'stream'\nimport type { ElevenLabsTTSOptions } from './types'\n\nconst DEFAULT_MAX_RETRY = 3\n\n/** One of the two ElevenLabs WebSockets, driven by ElevenLabsTTS */\nexport interface ElevenLabsWebSocket {\n speak(textStream: Readable): void\n cancel(): void\n destroy(): void\n}\n\n/** What a WebSocket hands back to the ElevenLabsTTS driving it */\nexport interface ElevenLabsWebSocketHost {\n audio(chunk: Buffer): void\n failed(texts: string[]): void\n log(...message: any[]): void\n}\n\nexport interface ElevenLabsConnectionOptions {\n url: string\n tts: ElevenLabsTTSOptions\n /** First message of every connection, which sets the session up */\n init: object\n keepAlive: { interval: number; message: object }\n host: ElevenLabsWebSocketHost\n /** The connection opened, after a drop when `reconnected` is true */\n onOpen(reconnected: boolean): void\n onMessage(message: any): void\n /** Every retry failed, and the connection waits for `connect()` */\n onFailed(): void\n}\n\n/**\n * The connection of one of the two WebSockets, reconnecting on its own.\n *\n * ElevenLabs opens a connection before it checks the key and the voice, then\n * closes it with an error. Such a connection opens fine, which would start the\n * retry count of ReconnectingSocket over each time and reconnect forever. So a\n * connection closed before it proved usable counts as a failed one, and the\n * connection gives up once they exceed `maxRetry` in a row.\n */\nexport class ElevenLabsConnection {\n private socket: ReconnectingSocket\n // Whether the current connection carried audio or lived long enough for a\n // keep-alive, which means the server accepted the key and the voice\n private proven = false\n private failures = 0\n\n constructor(private readonly options: ElevenLabsConnectionOptions) {\n this.socket = this.createSocket()\n }\n\n /** Opens the connection, again after giving up, and does nothing otherwise */\n connect() {\n this.socket.connect()\n }\n\n /**\n * Resolves to false when the connection gives up or is destroyed meanwhile,\n * and right away once it gave up, until the next `connect()`: the stream\n * that was speaking then has nothing left to wait for.\n */\n ready(): Promise<boolean> {\n if (this.socket.status === 'idle') return Promise.resolve(false)\n return this.socket.ready()\n }\n\n send(message: object): boolean {\n return this.socket.send(JSON.stringify(message))\n }\n\n /** Drops the connection, with the synthesis running on it, for a new one */\n restart() {\n this.socket.restart()\n }\n\n destroy() {\n this.socket.destroy()\n }\n\n /** The current connection works, so its next drop is not a failure */\n prove() {\n this.proven = true\n this.failures = 0\n }\n\n private createSocket() {\n const { tts, keepAlive, host } = this.options\n const socket = new ReconnectingSocket({\n url: this.options.url,\n clientOptions: { headers: { 'xi-api-key': tts.apiKey } },\n connectionTimeout: tts.connectionTimeout,\n retryDelay: tts.retryDelay,\n maxRetry: tts.maxRetry,\n handshake: {\n start: (ws) => ws.send(JSON.stringify(this.options.init)),\n },\n keepAlive: {\n interval: keepAlive.interval,\n message: () => {\n this.prove()\n return JSON.stringify(keepAlive.message)\n },\n },\n log: (...message) => host.log(...message),\n })\n\n socket.on('Open', ({ reconnected }) => {\n this.proven = false\n this.options.onOpen(reconnected)\n })\n\n socket.on('Message', (data) => {\n try {\n this.options.onMessage(JSON.parse(data.toString()))\n } catch (error: any) {\n host.log('message' in error ? error.message : error)\n host.log('Event data during error:', data.toString())\n }\n })\n\n socket.on('Close', () => {\n if (this.proven) return\n this.failures++\n if (this.failures > (tts.maxRetry ?? DEFAULT_MAX_RETRY)) this.giveUp()\n })\n\n socket.on('Failed', () => this.giveUp())\n return socket\n }\n\n private giveUp() {\n this.options.host.log('Max retries reached, giving up')\n this.failures = 0\n // An idle socket replaces the one that gave up, so the next utterance\n // connects again with a full retry budget\n this.socket.destroy()\n this.socket = this.createSocket()\n this.options.onFailed()\n }\n}\n","import type {\n TextToSpeechStreamRequestOutputFormat,\n VoiceSettings,\n} from '@elevenlabs/elevenlabs-js/api'\n\nexport type ElevenLabsModelId =\n // Text to Dialogue WebSocket\n | 'eleven_v4_turbo'\n | 'eleven_v4'\n | 'eleven_v3_conversational'\n | 'eleven_v3'\n // Text to Speech WebSocket\n | 'eleven_flash_v2_5'\n | 'eleven_multilingual_v2'\n | 'eleven_turbo_v2_5'\n\n/** Rates ElevenLabs streams raw PCM at. 44100 needs a Pro plan or above. */\nexport type ElevenLabsSampleRate =\n 8000 | 16000 | 22050 | 24000 | 32000 | 44100 | 48000\n\nexport interface ElevenLabsTTSOptions {\n apiKey: string\n voiceId: string\n modelId?: ElevenLabsModelId\n language?: string\n /**\n * Rate of the audio, in Hz, 48000 by default, the highest one offered. It\n * goes to the client as is, so the voice is never resampled on the server.\n */\n sampleRate?: ElevenLabsSampleRate\n /**\n * @deprecated Use `sampleRate`. The client only plays raw PCM, so only the\n * `pcm_*` formats work.\n */\n outputFormat?: TextToSpeechStreamRequestOutputFormat\n voiceSettings?: VoiceSettings\n connectionTimeout?: number\n retryDelay?: number\n maxRetry?: number\n}\n\nexport const DEFAULT_MODEL_ID = 'eleven_turbo_v2_5'\nexport const DEFAULT_SAMPLE_RATE: ElevenLabsSampleRate = 48000\n/** @deprecated Use `DEFAULT_SAMPLE_RATE` */\nexport const DEFAULT_OUTPUT_FORMAT = 'pcm_48000'\n\n/** The `output_format` asked of ElevenLabs */\nexport function getOutputFormat(\n options: ElevenLabsTTSOptions\n): TextToSpeechStreamRequestOutputFormat {\n if (options.sampleRate) return `pcm_${options.sampleRate}`\n return options.outputFormat ?? `pcm_${DEFAULT_SAMPLE_RATE}`\n}\n\n/** Rate of the audio ElevenLabs sends, read from its `output_format` */\nexport function getSampleRate(options: ElevenLabsTTSOptions): number {\n const match = /^pcm_(\\d+)$/.exec(getOutputFormat(options))\n return match ? Number(match[1]) : DEFAULT_SAMPLE_RATE\n}\n\n/** Eleven v3 and v4 only run on the Text to Dialogue WebSocket */\nexport function isDialogueModel(modelId: string): boolean {\n return modelId.startsWith('eleven_v3') || modelId.startsWith('eleven_v4')\n}\n\n// Text to Speech WebSocket\n\nexport type ElevenLabsWebSocketMessage =\n | ElevenLabsWebSocketAudioOutputMessage\n | ElevenLabsWebSocketFinalOutputMessage\n | ElevenLabsWebSocketErrorMessage\n\nexport interface ElevenLabsWebSocketAudioOutputMessage {\n audio: string\n normalizedAlignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n alignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n isFinal?: boolean | null\n}\n\nexport interface ElevenLabsWebSocketAudioOutputMessageAlignment {\n chars: string[]\n charStartTimesMs: number[]\n charDurationsMs: number[]\n}\n\nexport interface ElevenLabsWebSocketFinalOutputMessage {\n isFinal: boolean\n}\n\nexport interface ElevenLabsWebSocketErrorMessage {\n message: string\n error: string\n code: number\n}\n\n// Text to Dialogue WebSocket\n\nexport interface ElevenLabsDialogueClientMessage {\n voices?: string[]\n voice_settings?: { stability?: number }\n inputs?: ElevenLabsDialogueInput[]\n flush?: boolean\n keep_alive?: boolean\n}\n\nexport interface ElevenLabsDialogueInput {\n text: string\n voice_id: string\n new_turn?: boolean\n}\n\nexport type ElevenLabsDialogueMessage =\n | { audio: string }\n | { is_final_audio_for_turn: boolean }\n | { is_final: boolean }\n | ElevenLabsWebSocketErrorMessage\n","import { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsDialogueClientMessage,\n ElevenLabsDialogueMessage,\n ElevenLabsTTSOptions,\n getOutputFormat,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-dialogue/ttd-websocket\n\n// The server closes a connection that hears nothing for 20 seconds\nconst KEEP_ALIVE_INTERVAL = 15000\n\n// Punctuation closing a sentence, with the space that follows it. An ellipsis\n// is a pause within the sentence, which v4 performs, so it does not close one.\nconst SENTENCE_END = /(?:[!?]+|(?<!\\.)\\.(?!\\.))\\s+/g\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `streaming`: the text stream is still coming\n * - `draining`: the text stream ended, the audio of its flushes is coming\n */\ntype UtteranceState = 'idle' | 'streaming' | 'draining'\n\n/**\n * The Text to Dialogue WebSocket, the only one serving Eleven v3 and v4.\n *\n * The server holds text back until it has about 40 characters and 8 words,\n * which delays a short first sentence such as \"Hi there!\". Flushing at the end\n * of every sentence starts its audio right away.\n *\n * Nothing in the protocol stops a synthesis once it is sent, so cancelling\n * drops the connection and opens a fresh one. It opens in under 100 ms, while\n * the user is still speaking.\n */\nexport class DialogueWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private unflushed = false // Whether text was sent since the last flush\n private pendingFlushes = 0 // Flushes whose audio is not complete yet\n private newTurn = false // Whether the next input starts a new utterance\n private hasSpoken = false // Whether the connection received any text\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append('output_format', getOutputFormat(this.options))\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n // Register the voice. Dialogue models only read the stability.\n const stability = this.options.voiceSettings?.stability\n const init: ElevenLabsDialogueClientMessage = {\n voices: [this.options.voiceId],\n voice_settings: stability === undefined ? undefined : { stability },\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-dialogue/stream-input?${params.toString()}`,\n tts: options,\n init,\n keepAlive: {\n interval: KEEP_ALIVE_INTERVAL,\n message: { keep_alive: true },\n },\n host,\n onOpen: () => this.onOpen(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say, for the same\n // reason as on the Text to Speech WebSocket: a stream that never carries a\n // word must not cut off the sentence still being spoken.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'streaming'\n this.newTurn = true\n return true\n }\n\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendText(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.endText()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n if (this.textBuffer.trim()) {\n this.sendText(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.endText()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n private onOpen() {\n this.hasSpoken = false\n if (this.state === 'idle') return\n // The flushes in flight were lost with the previous connection, so the\n // utterance is spoken again from its start\n const text = this.textSent\n this.textSent = ''\n this.unflushed = false\n this.pendingFlushes = 0\n if (text.trim()) {\n this.log('Sending text again')\n this.sendText(text)\n }\n if (this.state === 'draining') this.flush()\n this.checkEnded()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle' ? '' : this.textSent + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.unflushed = false\n this.pendingFlushes = 0\n }\n\n /** The text stream is over: flushes what the server holds */\n private endText() {\n this.state = 'draining'\n this.flush()\n this.checkEnded()\n }\n\n private onMessage(message: ElevenLabsDialogueMessage) {\n if ('audio' in message && message.audio) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n } else {\n this.log(`Received audio chunk (${chunk.length} bytes)`)\n this.host.audio(chunk)\n }\n }\n if (\n 'is_final_audio_for_turn' in message &&\n message.is_final_audio_for_turn\n ) {\n this.pendingFlushes = Math.max(0, this.pendingFlushes - 1)\n this.checkEnded()\n }\n // Sent before the server closes the connection, for a wrong key, an\n // unknown voice or no dialogue session left\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n /** Sends text, flushing after each sentence it completes */\n private sendText(text: string) {\n let start = 0\n for (const match of text.matchAll(SENTENCE_END)) {\n const end = match.index! + match[0].length\n this.sendInput(text.slice(start, end), true)\n start = end\n }\n if (start < text.length) {\n this.sendInput(text.slice(start), false)\n }\n }\n\n private sendInput(text: string, flush: boolean) {\n this.textSent += text\n if (!text.trim()) return\n this.send({\n inputs: [\n {\n text,\n voice_id: this.options.voiceId,\n // Resets the prosody between two utterances\n new_turn: (this.newTurn && this.hasSpoken) || undefined,\n },\n ],\n flush: flush || undefined,\n })\n this.log(`Sent text${flush ? ' and flushed' : ''}: \"${text}\"`)\n this.newTurn = false\n this.hasSpoken = true\n if (flush) {\n this.pendingFlushes++\n this.unflushed = false\n } else {\n this.unflushed = true\n }\n }\n\n /** Generates the text held by the server, if any */\n private flush() {\n if (!this.unflushed) return\n this.send({ flush: true })\n this.pendingFlushes++\n this.unflushed = false\n this.log('Flushed text')\n }\n\n /** Ends the utterance once its text is over and all its audio came */\n private checkEnded() {\n if (this.state !== 'draining') return\n if (this.unflushed || this.pendingFlushes > 0) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n // Lost when the connection just dropped, which is fine: the reconnection\n // sends the utterance again from textSent\n private send(message: ElevenLabsDialogueClientMessage) {\n this.connection.send(message)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n","import type { VoiceSettings } from '@elevenlabs/elevenlabs-js/api'\nimport { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsTTSOptions,\n getOutputFormat,\n ElevenLabsWebSocketAudioOutputMessage,\n ElevenLabsWebSocketMessage,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-speech/v-1-text-to-speech-voice-id-stream-input\n\nconst WS_INACTIVITY_TIMEOUT = 180\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `speaking`: the text stream is still coming\n * - `flushed`: the text stream ended and was flushed, its audio is coming\n */\ntype UtteranceState = 'idle' | 'speaking' | 'flushed'\n\n/**\n * The Text to Speech WebSocket, for Flash, Multilingual v2 and Turbo v2.5.\n *\n * Its messages carry no utterance ID, and nothing in the protocol stops a\n * synthesis once it is sent. So cancelling drops the connection and opens a\n * fresh one, while the user is still speaking, and the audio of the cancelled\n * utterance is never heard of again.\n */\nexport class SpeechWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private heardUntil = 0 // Index in textSent where the audio received ends\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n // Build query params\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append('output_format', getOutputFormat(this.options))\n params.append('inactivity_timeout', WS_INACTIVITY_TIMEOUT.toString())\n const voiceSettings = toWireVoiceSettings(this.options.voiceSettings)\n if (voiceSettings) {\n params.append('voice_settings', JSON.stringify(voiceSettings))\n }\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-speech/${this.options.voiceId}/stream-input?${params.toString()}`,\n tts: options,\n // Initialization message with voice settings\n init: { text: ' ', voice_settings: voiceSettings },\n keepAlive: {\n interval: (WS_INACTIVITY_TIMEOUT - 1) * 1000,\n message: { text: ' ' },\n },\n host,\n onOpen: () => this.resume(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say.\n //\n // These fields track the utterance being spoken right now, and the\n // comparison between what was sent and what came back as audio is what\n // says where it is. Resetting them for a stream that never carries a word,\n // which is what an answer skipped by a tool or by onBeforeAnswer hands\n // over, loses the state of a sentence still being spoken and flushes it\n // for nothing.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'speaking'\n return true\n }\n\n // Forward text chunks coming from the caller to ElevenLabs\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendTranscript(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.flush()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n // Send last buffered text\n if (this.textBuffer.trim()) {\n this.sendTranscript(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.flush()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n /** Flushes buffered text and marks the end of the utterance */\n private flush() {\n this.state = 'flushed'\n this.connection.send({ text: ' ', flush: true })\n this.log('Flushed text')\n this.checkEnded()\n }\n\n /**\n * Sends the text of the utterance in flight again on a new connection. The\n * server lost what it was generating, so it goes on from the audio received.\n */\n private resume() {\n if (this.state === 'idle') return\n // Audio can stop inside a word, which goes again whole rather than from\n // its middle\n const wordStart = this.textSent.lastIndexOf(' ', this.heardUntil - 1) + 1\n const unheard = this.textSent.slice(wordStart)\n this.textSent = ''\n this.heardUntil = 0\n if (unheard.trim()) {\n this.log('Sending text chunks again')\n this.sendTranscript(unheard)\n }\n if (this.state === 'flushed') this.flush()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle'\n ? ''\n : this.textSent.slice(this.heardUntil) + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.heardUntil = 0\n }\n\n private onMessage(message: ElevenLabsWebSocketMessage) {\n if ('audio' in message && message.audio) {\n this.processAudioMessage(message)\n }\n if ('isFinal' in message && message.isFinal && this.state !== 'idle') {\n this.log('Audio ended')\n this.resetUtterance()\n }\n // Sent before the server closes the connection, for a wrong key or an\n // unknown voice\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n private processAudioMessage(message: ElevenLabsWebSocketAudioOutputMessage) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n return\n }\n\n // Check text of received audio chunk\n if (message.alignment) {\n const chunkText = message.alignment.chars.join('')\n\n // A connection carries one utterance at a time, since a cancel or a\n // new utterance opens another, so its audio is always wanted. The\n // alignment only tells how far the voice got, for a reconnection. It\n // follows the text sent, a few characters aside when the voice skips\n // or rewrites some, and is placed where it fits next.\n const end = findHeardEnd(this.textSent, this.heardUntil, chunkText)\n if (end === -1) {\n this.log(`Audio chunk off the text sent (\"${chunkText}\")`)\n } else {\n this.log(`Received audio chunk (\"${chunkText}\")`)\n this.heardUntil = end\n }\n }\n\n // Send to audio stream\n this.host.audio(chunk)\n\n // ElevenLabs doesn't send isFinal after a flush, so the end of the\n // utterance is when its audio covers all its text\n this.checkEnded()\n }\n\n private checkEnded() {\n if (this.state !== 'flushed') return\n if (this.textSent.slice(this.heardUntil).trim()) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n private sendTranscript(text: string) {\n // Counted as sent even when the connection just dropped, so the\n // reconnection sends it again\n this.textSent += text\n if (!this.connection.send({ text, try_trigger_generation: true })) return\n this.log(`Sent transcript: \"${text}\"`)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n\n/**\n * Where `heard` ends in `text`, at the first place from `from` where the text\n * goes on with it, or -1 when it does nowhere\n */\nfunction findHeardEnd(text: string, from: number, heard: string): number {\n for (let start = from; start < text.length; start++) {\n const end = endOfPrefix(text, start, heard)\n if (end !== -1) return end\n }\n return -1\n}\n\n/**\n * Where `heard` ends in `text`, read from `from` and compared without\n * whitespace, or -1 when the text does not go on with it there.\n */\nfunction endOfPrefix(text: string, from: number, heard: string): number {\n let index = from\n for (let i = 0; i < heard.length; i++) {\n if (isSpace(heard[i])) continue\n while (index < text.length && isSpace(text[index])) index++\n if (text[index] !== heard[i]) return -1\n index++\n }\n return index\n}\n\nfunction isSpace(char: string) {\n return char === ' ' || char === '\\t' || char === '\\n' || char === '\\r'\n}\n\n/**\n * Voice settings as the WebSocket reads them.\n *\n * The SDK types them in camel case and the WebSocket reads snake case, so\n * similarityBoost and useSpeakerBoost would be ignored if sent as they are.\n */\nfunction toWireVoiceSettings(settings?: VoiceSettings) {\n if (!settings) return undefined\n const { similarityBoost, useSpeakerBoost, ...rest } = settings\n return {\n similarity_boost: similarityBoost,\n use_speaker_boost: useSpeakerBoost,\n ...rest,\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACAA,IAAAA,iBAQO;;;ACRP,oBAAmC;AAInC,IAAM,oBAAoB;AAuCnB,IAAM,uBAAN,MAA2B;AAAA,EAOhC,YAA6B,SAAsC;AAAtC;AAH7B;AAAA;AAAA,SAAQ,SAAS;AACjB,SAAQ,WAAW;AAGjB,SAAK,SAAS,KAAK,aAAa;AAAA,EAClC;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,QAA0B;AACxB,QAAI,KAAK,OAAO,WAAW,OAAQ,QAAO,QAAQ,QAAQ,KAAK;AAC/D,WAAO,KAAK,OAAO,MAAM;AAAA,EAC3B;AAAA,EAEA,KAAK,SAA0B;AAC7B,WAAO,KAAK,OAAO,KAAK,KAAK,UAAU,OAAO,CAAC;AAAA,EACjD;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA,EAEA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,SAAS;AACd,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,eAAe;AACrB,UAAM,EAAE,KAAK,WAAW,KAAK,IAAI,KAAK;AACtC,UAAM,SAAS,IAAI,iCAAmB;AAAA,MACpC,KAAK,KAAK,QAAQ;AAAA,MAClB,eAAe,EAAE,SAAS,EAAE,cAAc,IAAI,OAAO,EAAE;AAAA,MACvD,mBAAmB,IAAI;AAAA,MACvB,YAAY,IAAI;AAAA,MAChB,UAAU,IAAI;AAAA,MACd,WAAW;AAAA,QACT,OAAO,CAAC,OAAO,GAAG,KAAK,KAAK,UAAU,KAAK,QAAQ,IAAI,CAAC;AAAA,MAC1D;AAAA,MACA,WAAW;AAAA,QACT,UAAU,UAAU;AAAA,QACpB,SAAS,MAAM;AACb,eAAK,MAAM;AACX,iBAAO,KAAK,UAAU,UAAU,OAAO;AAAA,QACzC;AAAA,MACF;AAAA,MACA,KAAK,IAAI,YAAY,KAAK,IAAI,GAAG,OAAO;AAAA,IAC1C,CAAC;AAED,WAAO,GAAG,QAAQ,CAAC,EAAE,YAAY,MAAM;AACrC,WAAK,SAAS;AACd,WAAK,QAAQ,OAAO,WAAW;AAAA,IACjC,CAAC;AAED,WAAO,GAAG,WAAW,CAAC,SAAS;AAC7B,UAAI;AACF,aAAK,QAAQ,UAAU,KAAK,MAAM,KAAK,SAAS,CAAC,CAAC;AAAA,MACpD,SAAS,OAAY;AACnB,aAAK,IAAI,aAAa,QAAQ,MAAM,UAAU,KAAK;AACnD,aAAK,IAAI,4BAA4B,KAAK,SAAS,CAAC;AAAA,MACtD;AAAA,IACF,CAAC;AAED,WAAO,GAAG,SAAS,MAAM;AACvB,UAAI,KAAK,OAAQ;AACjB,WAAK;AACL,UAAI,KAAK,YAAY,IAAI,YAAY,mBAAoB,MAAK,OAAO;AAAA,IACvE,CAAC;AAED,WAAO,GAAG,UAAU,MAAM,KAAK,OAAO,CAAC;AACvC,WAAO;AAAA,EACT;AAAA,EAEQ,SAAS;AACf,SAAK,QAAQ,KAAK,IAAI,gCAAgC;AACtD,SAAK,WAAW;AAGhB,SAAK,OAAO,QAAQ;AACpB,SAAK,SAAS,KAAK,aAAa;AAChC,SAAK,QAAQ,SAAS;AAAA,EACxB;AACF;;;ACrGO,IAAM,mBAAmB;AACzB,IAAM,sBAA4C;AAElD,IAAM,wBAAwB;AAG9B,SAAS,gBACd,SACuC;AACvC,MAAI,QAAQ,WAAY,QAAO,OAAO,QAAQ,UAAU;AACxD,SAAO,QAAQ,gBAAgB,OAAO,mBAAmB;AAC3D;AAGO,SAAS,cAAc,SAAuC;AACnE,QAAM,QAAQ,cAAc,KAAK,gBAAgB,OAAO,CAAC;AACzD,SAAO,QAAQ,OAAO,MAAM,CAAC,CAAC,IAAI;AACpC;AAGO,SAAS,gBAAgB,SAA0B;AACxD,SAAO,QAAQ,WAAW,WAAW,KAAK,QAAQ,WAAW,WAAW;AAC1E;;;AC9CA,IAAM,sBAAsB;AAI5B,IAAM,eAAe;AAsBd,IAAM,oBAAN,MAAuD;AAAA;AAAA,EAgB5D,YACmB,SACA,MACjB;AAFiB;AACA;AAhBnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,YAAY;AACpB;AAAA,SAAQ,iBAAiB;AACzB;AAAA,SAAQ,UAAU;AAClB;AAAA,SAAQ,YAAY;AAMlB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO,OAAO,iBAAiB,gBAAgB,KAAK,OAAO,CAAC;AAC5D,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAGA,UAAM,YAAY,KAAK,QAAQ,eAAe;AAC9C,UAAM,OAAwC;AAAA,MAC5C,QAAQ,CAAC,KAAK,QAAQ,OAAO;AAAA,MAC7B,gBAAgB,cAAc,SAAY,SAAY,EAAE,UAAU;AAAA,IACpE;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,4DAA4D,OAAO,SAAS,CAAC;AAAA,MAClF,KAAK;AAAA,MACL;AAAA,MACA,WAAW;AAAA,QACT,UAAU;AAAA,QACV,SAAS,EAAE,YAAY,KAAK;AAAA,MAC9B;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAKd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,WAAK,UAAU;AACf,aAAO;AAAA,IACT;AAEA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,SAAS,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AACtD,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAClC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,SAAS,KAAK,aAAa,GAAG;AAAA,MACrC;AACA,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEQ,SAAS;AACf,SAAK,YAAY;AACjB,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,OAAO,KAAK;AAClB,SAAK,WAAW;AAChB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AACtB,QAAI,KAAK,KAAK,GAAG;AACf,WAAK,IAAI,oBAAoB;AAC7B,WAAK,SAAS,IAAI;AAAA,IACpB;AACA,QAAI,KAAK,UAAU,WAAY,MAAK,MAAM;AAC1C,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SAAS,KAAK,KAAK,WAAW,KAAK;AACpD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AAAA,EACxB;AAAA;AAAA,EAGQ,UAAU;AAChB,SAAK,QAAQ;AACb,SAAK,MAAM;AACX,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,UAAU,SAAoC;AACpD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,YAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,WAAK,WAAW,MAAM;AACtB,UAAI,KAAK,UAAU,QAAQ;AACzB,aAAK,IAAI,6CAA6C;AAAA,MACxD,OAAO;AACL,aAAK,IAAI,yBAAyB,MAAM,MAAM,SAAS;AACvD,aAAK,KAAK,MAAM,KAAK;AAAA,MACvB;AAAA,IACF;AACA,QACE,6BAA6B,WAC7B,QAAQ,yBACR;AACA,WAAK,iBAAiB,KAAK,IAAI,GAAG,KAAK,iBAAiB,CAAC;AACzD,WAAK,WAAW;AAAA,IAClB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA;AAAA,EAGQ,SAAS,MAAc;AAC7B,QAAI,QAAQ;AACZ,eAAW,SAAS,KAAK,SAAS,YAAY,GAAG;AAC/C,YAAM,MAAM,MAAM,QAAS,MAAM,CAAC,EAAE;AACpC,WAAK,UAAU,KAAK,MAAM,OAAO,GAAG,GAAG,IAAI;AAC3C,cAAQ;AAAA,IACV;AACA,QAAI,QAAQ,KAAK,QAAQ;AACvB,WAAK,UAAU,KAAK,MAAM,KAAK,GAAG,KAAK;AAAA,IACzC;AAAA,EACF;AAAA,EAEQ,UAAU,MAAc,OAAgB;AAC9C,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,KAAK,EAAG;AAClB,SAAK,KAAK;AAAA,MACR,QAAQ;AAAA,QACN;AAAA,UACE;AAAA,UACA,UAAU,KAAK,QAAQ;AAAA;AAAA,UAEvB,UAAW,KAAK,WAAW,KAAK,aAAc;AAAA,QAChD;AAAA,MACF;AAAA,MACA,OAAO,SAAS;AAAA,IAClB,CAAC;AACD,SAAK,IAAI,YAAY,QAAQ,iBAAiB,EAAE,MAAM,IAAI,GAAG;AAC7D,SAAK,UAAU;AACf,SAAK,YAAY;AACjB,QAAI,OAAO;AACT,WAAK;AACL,WAAK,YAAY;AAAA,IACnB,OAAO;AACL,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGQ,QAAQ;AACd,QAAI,CAAC,KAAK,UAAW;AACrB,SAAK,KAAK,EAAE,OAAO,KAAK,CAAC;AACzB,SAAK;AACL,SAAK,YAAY;AACjB,SAAK,IAAI,cAAc;AAAA,EACzB;AAAA;AAAA,EAGQ,aAAa;AACnB,QAAI,KAAK,UAAU,WAAY;AAC/B,QAAI,KAAK,aAAa,KAAK,iBAAiB,EAAG;AAC/C,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA;AAAA,EAIQ,KAAK,SAA0C;AACrD,SAAK,WAAW,KAAK,OAAO;AAAA,EAC9B;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;;;ACxRA,IAAM,wBAAwB;AAmBvB,IAAM,kBAAN,MAAqD;AAAA;AAAA,EAa1D,YACmB,SACA,MACjB;AAFiB;AACA;AAbnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,aAAa;AAOnB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO,OAAO,iBAAiB,gBAAgB,KAAK,OAAO,CAAC;AAC5D,WAAO,OAAO,sBAAsB,sBAAsB,SAAS,CAAC;AACpE,UAAM,gBAAgB,oBAAoB,KAAK,QAAQ,aAAa;AACpE,QAAI,eAAe;AACjB,aAAO,OAAO,kBAAkB,KAAK,UAAU,aAAa,CAAC;AAAA,IAC/D;AACA,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,6CAA6C,KAAK,QAAQ,OAAO,iBAAiB,OAAO,SAAS,CAAC;AAAA,MACxG,KAAK;AAAA;AAAA,MAEL,MAAM,EAAE,MAAM,KAAK,gBAAgB,cAAc;AAAA,MACjD,WAAW;AAAA,QACT,WAAW,wBAAwB,KAAK;AAAA,QACxC,SAAS,EAAE,MAAM,IAAI;AAAA,MACvB;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAUd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,aAAO;AAAA,IACT;AAGA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,eAAe,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AAC5D,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAElC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,eAAe,KAAK,aAAa,GAAG;AAAA,MAC3C;AACA,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA;AAAA,EAGQ,QAAQ;AACd,SAAK,QAAQ;AACb,SAAK,WAAW,KAAK,EAAE,MAAM,KAAK,OAAO,KAAK,CAAC;AAC/C,SAAK,IAAI,cAAc;AACvB,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMQ,SAAS;AACf,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,YAAY,KAAK,SAAS,YAAY,KAAK,KAAK,aAAa,CAAC,IAAI;AACxE,UAAM,UAAU,KAAK,SAAS,MAAM,SAAS;AAC7C,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,QAAI,QAAQ,KAAK,GAAG;AAClB,WAAK,IAAI,2BAA2B;AACpC,WAAK,eAAe,OAAO;AAAA,IAC7B;AACA,QAAI,KAAK,UAAU,UAAW,MAAK,MAAM;AAAA,EAC3C;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SACX,KACA,KAAK,SAAS,MAAM,KAAK,UAAU,IAAI,KAAK;AAClD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,aAAa;AAAA,EACpB;AAAA,EAEQ,UAAU,SAAqC;AACrD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,WAAK,oBAAoB,OAAO;AAAA,IAClC;AACA,QAAI,aAAa,WAAW,QAAQ,WAAW,KAAK,UAAU,QAAQ;AACpE,WAAK,IAAI,aAAa;AACtB,WAAK,eAAe;AAAA,IACtB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA,EAEQ,oBAAoB,SAAgD;AAC1E,UAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,SAAK,WAAW,MAAM;AACtB,QAAI,KAAK,UAAU,QAAQ;AACzB,WAAK,IAAI,6CAA6C;AACtD;AAAA,IACF;AAGA,QAAI,QAAQ,WAAW;AACrB,YAAM,YAAY,QAAQ,UAAU,MAAM,KAAK,EAAE;AAOjD,YAAM,MAAM,aAAa,KAAK,UAAU,KAAK,YAAY,SAAS;AAClE,UAAI,QAAQ,IAAI;AACd,aAAK,IAAI,mCAAmC,SAAS,IAAI;AAAA,MAC3D,OAAO;AACL,aAAK,IAAI,0BAA0B,SAAS,IAAI;AAChD,aAAK,aAAa;AAAA,MACpB;AAAA,IACF;AAGA,SAAK,KAAK,MAAM,KAAK;AAIrB,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,aAAa;AACnB,QAAI,KAAK,UAAU,UAAW;AAC9B,QAAI,KAAK,SAAS,MAAM,KAAK,UAAU,EAAE,KAAK,EAAG;AACjD,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA,EAEQ,eAAe,MAAc;AAGnC,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,WAAW,KAAK,EAAE,MAAM,wBAAwB,KAAK,CAAC,EAAG;AACnE,SAAK,IAAI,qBAAqB,IAAI,GAAG;AAAA,EACvC;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;AAMA,SAAS,aAAa,MAAc,MAAc,OAAuB;AACvE,WAAS,QAAQ,MAAM,QAAQ,KAAK,QAAQ,SAAS;AACnD,UAAM,MAAM,YAAY,MAAM,OAAO,KAAK;AAC1C,QAAI,QAAQ,GAAI,QAAO;AAAA,EACzB;AACA,SAAO;AACT;AAMA,SAAS,YAAY,MAAc,MAAc,OAAuB;AACtE,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,QAAI,QAAQ,MAAM,CAAC,CAAC,EAAG;AACvB,WAAO,QAAQ,KAAK,UAAU,QAAQ,KAAK,KAAK,CAAC,EAAG;AACpD,QAAI,KAAK,KAAK,MAAM,MAAM,CAAC,EAAG,QAAO;AACrC;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,QAAQ,MAAc;AAC7B,SAAO,SAAS,OAAO,SAAS,OAAQ,SAAS,QAAQ,SAAS;AACpE;AAQA,SAAS,oBAAoB,UAA0B;AACrD,MAAI,CAAC,SAAU,QAAO;AACtB,QAAM,EAAE,iBAAiB,iBAAiB,GAAG,KAAK,IAAI;AACtD,SAAO;AAAA,IACL,kBAAkB;AAAA,IAClB,mBAAmB;AAAA,IACnB,GAAG;AAAA,EACL;AACF;;;AJlSA,IAAM,oBAAoB,CAAC,UAAU,aAAa,cAAc;AASzD,IAAM,gBAAN,cAA4B,mBAAI;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOrC,OAAO,WAAW,SAGU;AAC1B,eAAO;AAAA,MACL,CAAC,YAAY;AAAA,MACb,QAAQ;AAAA,MACR,YAAY;AACV,cAAM,WAAW,MAAM,MAAM,uCAAuC;AAAA,UAClE,SAAS,EAAE,cAAc,QAAQ,OAAO;AAAA,QAC1C,CAAC;AACD,YAAI,CAAC,SAAS,IAAI;AAChB,gBAAM,IAAI,MAAM,uBAAuB,SAAS,MAAM,EAAE;AAAA,QAC1D;AACA,cAAM,SAAU,MAAM,SAAS,KAAK;AAOpC,eAAO,OACJ,OAAO,CAAC,UAAU,MAAM,qBAAqB,EAC7C,IAAI,CAAC,WAAW;AAAA,UACf,IAAI,MAAM;AAAA,UACV,MAAM,MAAM;AAAA,UACZ,aAAa,MAAM;AAAA,UACnB,WAAW,MAAM,WAAW,IAAI,CAAC,EAAE,YAAY,MAAM,WAAW;AAAA,QAClE,EAAE;AAAA,MACN;AAAA,MACA,QAAQ;AAAA,IACV;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAWA,OAAO,WAAW,SAIU;AAC1B,eAAO;AAAA,MACL,CAAC,YAAY;AAAA,MACb,QAAQ;AAAA,MACR,MAAM,YAAY,QAAQ,MAAM;AAAA,MAChC,QAAQ;AAAA,IACV,EAAE,KAAK,CAAC,eAAW,kCAAkB,QAAQ,QAAQ,QAAQ,CAAC;AAAA,EAChE;AAAA,EAKA,YAAY,SAA+B;AACzC,UAAM;AACN,SAAK,OAAO,cAAc,OAAO;AACjC,UAAM,OAAO;AAAA,MACX,OAAO,CAAC,UAAkB,KAAK,KAAK,SAAS,KAAK;AAAA,MAClD,QAAQ,CAAC,UAAoB,KAAK,KAAK,UAAU,KAAK;AAAA,MACtD,KAAK,IAAI,YAAmB,KAAK,IAAI,GAAG,OAAO;AAAA,IACjD;AACA,SAAK,YAAY,gBAAgB,QAAQ,WAAW,gBAAgB,IAChE,IAAI,kBAAkB,SAAS,IAAI,IACnC,IAAI,gBAAgB,SAAS,IAAI;AAAA,EACvC;AAAA,EAEA,IAAI,aAAa;AACf,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAM,YAAsB;AAC1B,SAAK,UAAU,MAAM,UAAU;AAAA,EACjC;AAAA,EAEA,SAAS;AACP,SAAK,UAAU,OAAO;AAAA,EACxB;AAAA,EAEA,UAAU;AACR,SAAK,UAAU,QAAQ;AACvB,UAAM,QAAQ;AAAA,EAChB;AACF;AAGA,eAAe,YAAY,QAAyC;AAClE,QAAM,SAAyB,CAAC;AAChC,MAAI;AACJ,KAAG;AACD,UAAM,SAAS,IAAI,gBAAgB,EAAE,WAAW,MAAM,CAAC;AACvD,QAAI,UAAW,QAAO,IAAI,mBAAmB,SAAS;AACtD,UAAM,WAAW,MAAM;AAAA,MACrB,uCAAuC,MAAM;AAAA,MAC7C,EAAE,SAAS,EAAE,cAAc,OAAO,EAAE;AAAA,IACtC;AACA,QAAI,CAAC,SAAS,GAAI,OAAM,IAAI,MAAM,uBAAuB,SAAS,MAAM,EAAE;AAC1E,UAAM,OAAQ,MAAM,SAAS,KAAK;AAKlC,WAAO,KAAK,GAAG,KAAK,OAAO,IAAI,cAAc,CAAC;AAC9C,gBAAY,KAAK,WAAY,KAAK,mBAAmB,SAAa;AAAA,EACpE,SAAS;AACT,SAAO;AACT;AAEA,SAAS,eAAe,OAAsC;AAC5D,QAAM,YAAY,MAAM,sBAAsB,CAAC,GAC5C,IAAI,CAAC,EAAE,QAAQ,SAAS,MAAM,UAAU,QAAQ,EAChD,OAAO,CAAC,SAAyB,CAAC,CAAC,IAAI;AAC1C,QAAM,QAAQ,MAAM,QAAQ;AAC5B,QAAM,YAAY,SAAS,SAAS,WAAW,QAAQ,CAAC,KAAK,IAAI;AACjE,SAAO;AAAA,IACL,IAAI,MAAM;AAAA,IACV,MAAM,MAAM;AAAA,IACZ,WAAW,aAAa,CAAC,GAAG,IAAI,IAAI,SAAS,CAAC;AAAA,IAC9C,YAAQ,4BAAY,MAAM,QAAQ,MAAM;AAAA,IACxC,aAAa,MAAM,eAAe;AAAA,IAClC,YAAY,MAAM,eAAe;AAAA,IACjC,QAAQ,kBAAkB,SAAS,MAAM,YAAY,EAAE;AAAA,EACzD;AACF;","names":["import_server"]}
|
package/dist/index.mjs
CHANGED
|
@@ -1,5 +1,11 @@
|
|
|
1
1
|
// src/ElevenLabsTTS.ts
|
|
2
|
-
import {
|
|
2
|
+
import {
|
|
3
|
+
cachedModels,
|
|
4
|
+
cachedVoices,
|
|
5
|
+
TTS,
|
|
6
|
+
voiceGender,
|
|
7
|
+
voicesForLanguage
|
|
8
|
+
} from "@micdrop/server";
|
|
3
9
|
|
|
4
10
|
// src/ElevenLabsWebSocket.ts
|
|
5
11
|
import { ReconnectingSocket } from "@micdrop/server";
|
|
@@ -92,7 +98,16 @@ var ElevenLabsConnection = class {
|
|
|
92
98
|
|
|
93
99
|
// src/types.ts
|
|
94
100
|
var DEFAULT_MODEL_ID = "eleven_turbo_v2_5";
|
|
95
|
-
var
|
|
101
|
+
var DEFAULT_SAMPLE_RATE = 48e3;
|
|
102
|
+
var DEFAULT_OUTPUT_FORMAT = "pcm_48000";
|
|
103
|
+
function getOutputFormat(options) {
|
|
104
|
+
if (options.sampleRate) return `pcm_${options.sampleRate}`;
|
|
105
|
+
return options.outputFormat ?? `pcm_${DEFAULT_SAMPLE_RATE}`;
|
|
106
|
+
}
|
|
107
|
+
function getSampleRate(options) {
|
|
108
|
+
const match = /^pcm_(\d+)$/.exec(getOutputFormat(options));
|
|
109
|
+
return match ? Number(match[1]) : DEFAULT_SAMPLE_RATE;
|
|
110
|
+
}
|
|
96
111
|
function isDialogueModel(modelId) {
|
|
97
112
|
return modelId.startsWith("eleven_v3") || modelId.startsWith("eleven_v4");
|
|
98
113
|
}
|
|
@@ -125,10 +140,7 @@ var DialogueWebSocket = class {
|
|
|
125
140
|
this.hasSpoken = false;
|
|
126
141
|
const params = new URLSearchParams();
|
|
127
142
|
params.append("model_id", this.options.modelId ?? DEFAULT_MODEL_ID);
|
|
128
|
-
params.append(
|
|
129
|
-
"output_format",
|
|
130
|
-
this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT
|
|
131
|
-
);
|
|
143
|
+
params.append("output_format", getOutputFormat(this.options));
|
|
132
144
|
if (this.options.language) {
|
|
133
145
|
params.append("language_code", this.options.language);
|
|
134
146
|
}
|
|
@@ -344,10 +356,7 @@ var SpeechWebSocket = class {
|
|
|
344
356
|
this.heardUntil = 0;
|
|
345
357
|
const params = new URLSearchParams();
|
|
346
358
|
params.append("model_id", this.options.modelId ?? DEFAULT_MODEL_ID);
|
|
347
|
-
params.append(
|
|
348
|
-
"output_format",
|
|
349
|
-
this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT
|
|
350
|
-
);
|
|
359
|
+
params.append("output_format", getOutputFormat(this.options));
|
|
351
360
|
params.append("inactivity_timeout", WS_INACTIVITY_TIMEOUT.toString());
|
|
352
361
|
const voiceSettings = toWireVoiceSettings(this.options.voiceSettings);
|
|
353
362
|
if (voiceSettings) {
|
|
@@ -542,9 +551,56 @@ function toWireVoiceSettings(settings) {
|
|
|
542
551
|
}
|
|
543
552
|
|
|
544
553
|
// src/ElevenLabsTTS.ts
|
|
554
|
+
var CUSTOM_CATEGORIES = ["cloned", "generated", "professional"];
|
|
545
555
|
var ElevenLabsTTS = class extends TTS {
|
|
556
|
+
/**
|
|
557
|
+
* Lists the models that speak, with the languages each one reads.
|
|
558
|
+
*
|
|
559
|
+
* The list is fetched once and shared by every call of the process for a
|
|
560
|
+
* day. `refresh` fetches it again.
|
|
561
|
+
*/
|
|
562
|
+
static listModels(options) {
|
|
563
|
+
return cachedModels(
|
|
564
|
+
["elevenlabs"],
|
|
565
|
+
options.apiKey,
|
|
566
|
+
async () => {
|
|
567
|
+
const response = await fetch("https://api.elevenlabs.io/v1/models", {
|
|
568
|
+
headers: { "xi-api-key": options.apiKey }
|
|
569
|
+
});
|
|
570
|
+
if (!response.ok) {
|
|
571
|
+
throw new Error(`ElevenLabs answered ${response.status}`);
|
|
572
|
+
}
|
|
573
|
+
const models = await response.json();
|
|
574
|
+
return models.filter((model) => model.can_do_text_to_speech).map((model) => ({
|
|
575
|
+
id: model.model_id,
|
|
576
|
+
name: model.name,
|
|
577
|
+
description: model.description,
|
|
578
|
+
languages: model.languages?.map(({ language_id }) => language_id)
|
|
579
|
+
}));
|
|
580
|
+
},
|
|
581
|
+
options.refresh
|
|
582
|
+
);
|
|
583
|
+
}
|
|
584
|
+
/**
|
|
585
|
+
* Lists the voices the account can speak with: the premade ones, and the
|
|
586
|
+
* ones cloned, designed or added from the library.
|
|
587
|
+
*
|
|
588
|
+
* `language`, such as `fr` or `fr-FR`, keeps the voices that speak it.
|
|
589
|
+
*
|
|
590
|
+
* The list is fetched once and shared by every call of the process for a
|
|
591
|
+
* day. `refresh` fetches it again, for a voice added a moment ago.
|
|
592
|
+
*/
|
|
593
|
+
static listVoices(options) {
|
|
594
|
+
return cachedVoices(
|
|
595
|
+
["elevenlabs"],
|
|
596
|
+
options.apiKey,
|
|
597
|
+
() => fetchVoices(options.apiKey),
|
|
598
|
+
options.refresh
|
|
599
|
+
).then((voices) => voicesForLanguage(voices, options.language));
|
|
600
|
+
}
|
|
546
601
|
constructor(options) {
|
|
547
602
|
super();
|
|
603
|
+
this.rate = getSampleRate(options);
|
|
548
604
|
const host = {
|
|
549
605
|
audio: (chunk) => this.emit("Audio", chunk),
|
|
550
606
|
failed: (texts) => this.emit("Failed", texts),
|
|
@@ -552,6 +608,9 @@ var ElevenLabsTTS = class extends TTS {
|
|
|
552
608
|
};
|
|
553
609
|
this.webSocket = isDialogueModel(options.modelId ?? DEFAULT_MODEL_ID) ? new DialogueWebSocket(options, host) : new SpeechWebSocket(options, host);
|
|
554
610
|
}
|
|
611
|
+
get sampleRate() {
|
|
612
|
+
return this.rate;
|
|
613
|
+
}
|
|
555
614
|
speak(textStream) {
|
|
556
615
|
this.webSocket.speak(textStream);
|
|
557
616
|
}
|
|
@@ -563,10 +622,44 @@ var ElevenLabsTTS = class extends TTS {
|
|
|
563
622
|
super.destroy();
|
|
564
623
|
}
|
|
565
624
|
};
|
|
625
|
+
async function fetchVoices(apiKey) {
|
|
626
|
+
const voices = [];
|
|
627
|
+
let pageToken;
|
|
628
|
+
do {
|
|
629
|
+
const params = new URLSearchParams({ page_size: "100" });
|
|
630
|
+
if (pageToken) params.set("next_page_token", pageToken);
|
|
631
|
+
const response = await fetch(
|
|
632
|
+
`https://api.elevenlabs.io/v2/voices?${params}`,
|
|
633
|
+
{ headers: { "xi-api-key": apiKey } }
|
|
634
|
+
);
|
|
635
|
+
if (!response.ok) throw new Error(`ElevenLabs answered ${response.status}`);
|
|
636
|
+
const page = await response.json();
|
|
637
|
+
voices.push(...page.voices.map(toMicdropVoice));
|
|
638
|
+
pageToken = page.has_more ? page.next_page_token ?? void 0 : void 0;
|
|
639
|
+
} while (pageToken);
|
|
640
|
+
return voices;
|
|
641
|
+
}
|
|
642
|
+
function toMicdropVoice(voice) {
|
|
643
|
+
const verified = (voice.verified_languages ?? []).map(({ locale, language }) => locale || language).filter((code) => !!code);
|
|
644
|
+
const label = voice.labels?.language;
|
|
645
|
+
const languages = verified.length ? verified : label ? [label] : void 0;
|
|
646
|
+
return {
|
|
647
|
+
id: voice.voice_id,
|
|
648
|
+
name: voice.name,
|
|
649
|
+
languages: languages && [...new Set(languages)],
|
|
650
|
+
gender: voiceGender(voice.labels?.gender),
|
|
651
|
+
description: voice.description || void 0,
|
|
652
|
+
previewUrl: voice.preview_url || void 0,
|
|
653
|
+
custom: CUSTOM_CATEGORIES.includes(voice.category ?? "")
|
|
654
|
+
};
|
|
655
|
+
}
|
|
566
656
|
export {
|
|
567
657
|
DEFAULT_MODEL_ID,
|
|
568
658
|
DEFAULT_OUTPUT_FORMAT,
|
|
659
|
+
DEFAULT_SAMPLE_RATE,
|
|
569
660
|
ElevenLabsTTS,
|
|
661
|
+
getOutputFormat,
|
|
662
|
+
getSampleRate,
|
|
570
663
|
isDialogueModel
|
|
571
664
|
};
|
|
572
665
|
//# sourceMappingURL=index.mjs.map
|
package/dist/index.mjs.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/ElevenLabsTTS.ts","../src/ElevenLabsWebSocket.ts","../src/types.ts","../src/DialogueWebSocket.ts","../src/SpeechWebSocket.ts"],"sourcesContent":["import { TTS } from '@micdrop/server'\nimport { Readable } from 'stream'\nimport { DialogueWebSocket } from './DialogueWebSocket'\nimport { ElevenLabsWebSocket } from './ElevenLabsWebSocket'\nimport { SpeechWebSocket } from './SpeechWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsTTSOptions,\n isDialogueModel,\n} from './types'\n\n/**\n * ElevenLabs text to speech, streamed over the WebSocket the model runs on.\n *\n * Eleven v3 and v4 only answer on the Text to Dialogue WebSocket, the other\n * models on the Text to Speech WebSocket. The two speak different protocols,\n * so the model picks the one this class drives.\n */\nexport class ElevenLabsTTS extends TTS {\n private readonly webSocket: ElevenLabsWebSocket\n\n constructor(options: ElevenLabsTTSOptions) {\n super()\n const host = {\n audio: (chunk: Buffer) => this.emit('Audio', chunk),\n failed: (texts: string[]) => this.emit('Failed', texts),\n log: (...message: any[]) => this.log(...message),\n }\n this.webSocket = isDialogueModel(options.modelId ?? DEFAULT_MODEL_ID)\n ? new DialogueWebSocket(options, host)\n : new SpeechWebSocket(options, host)\n }\n\n speak(textStream: Readable) {\n this.webSocket.speak(textStream)\n }\n\n cancel() {\n this.webSocket.cancel()\n }\n\n destroy() {\n this.webSocket.destroy()\n super.destroy()\n }\n}\n","import { ReconnectingSocket } from '@micdrop/server'\nimport type { Readable } from 'stream'\nimport type { ElevenLabsTTSOptions } from './types'\n\nconst DEFAULT_MAX_RETRY = 3\n\n/** One of the two ElevenLabs WebSockets, driven by ElevenLabsTTS */\nexport interface ElevenLabsWebSocket {\n speak(textStream: Readable): void\n cancel(): void\n destroy(): void\n}\n\n/** What a WebSocket hands back to the ElevenLabsTTS driving it */\nexport interface ElevenLabsWebSocketHost {\n audio(chunk: Buffer): void\n failed(texts: string[]): void\n log(...message: any[]): void\n}\n\nexport interface ElevenLabsConnectionOptions {\n url: string\n tts: ElevenLabsTTSOptions\n /** First message of every connection, which sets the session up */\n init: object\n keepAlive: { interval: number; message: object }\n host: ElevenLabsWebSocketHost\n /** The connection opened, after a drop when `reconnected` is true */\n onOpen(reconnected: boolean): void\n onMessage(message: any): void\n /** Every retry failed, and the connection waits for `connect()` */\n onFailed(): void\n}\n\n/**\n * The connection of one of the two WebSockets, reconnecting on its own.\n *\n * ElevenLabs opens a connection before it checks the key and the voice, then\n * closes it with an error. Such a connection opens fine, which would start the\n * retry count of ReconnectingSocket over each time and reconnect forever. So a\n * connection closed before it proved usable counts as a failed one, and the\n * connection gives up once they exceed `maxRetry` in a row.\n */\nexport class ElevenLabsConnection {\n private socket: ReconnectingSocket\n // Whether the current connection carried audio or lived long enough for a\n // keep-alive, which means the server accepted the key and the voice\n private proven = false\n private failures = 0\n\n constructor(private readonly options: ElevenLabsConnectionOptions) {\n this.socket = this.createSocket()\n }\n\n /** Opens the connection, again after giving up, and does nothing otherwise */\n connect() {\n this.socket.connect()\n }\n\n /**\n * Resolves to false when the connection gives up or is destroyed meanwhile,\n * and right away once it gave up, until the next `connect()`: the stream\n * that was speaking then has nothing left to wait for.\n */\n ready(): Promise<boolean> {\n if (this.socket.status === 'idle') return Promise.resolve(false)\n return this.socket.ready()\n }\n\n send(message: object): boolean {\n return this.socket.send(JSON.stringify(message))\n }\n\n /** Drops the connection, with the synthesis running on it, for a new one */\n restart() {\n this.socket.restart()\n }\n\n destroy() {\n this.socket.destroy()\n }\n\n /** The current connection works, so its next drop is not a failure */\n prove() {\n this.proven = true\n this.failures = 0\n }\n\n private createSocket() {\n const { tts, keepAlive, host } = this.options\n const socket = new ReconnectingSocket({\n url: this.options.url,\n clientOptions: { headers: { 'xi-api-key': tts.apiKey } },\n connectionTimeout: tts.connectionTimeout,\n retryDelay: tts.retryDelay,\n maxRetry: tts.maxRetry,\n handshake: {\n start: (ws) => ws.send(JSON.stringify(this.options.init)),\n },\n keepAlive: {\n interval: keepAlive.interval,\n message: () => {\n this.prove()\n return JSON.stringify(keepAlive.message)\n },\n },\n log: (...message) => host.log(...message),\n })\n\n socket.on('Open', ({ reconnected }) => {\n this.proven = false\n this.options.onOpen(reconnected)\n })\n\n socket.on('Message', (data) => {\n try {\n this.options.onMessage(JSON.parse(data.toString()))\n } catch (error: any) {\n host.log('message' in error ? error.message : error)\n host.log('Event data during error:', data.toString())\n }\n })\n\n socket.on('Close', () => {\n if (this.proven) return\n this.failures++\n if (this.failures > (tts.maxRetry ?? DEFAULT_MAX_RETRY)) this.giveUp()\n })\n\n socket.on('Failed', () => this.giveUp())\n return socket\n }\n\n private giveUp() {\n this.options.host.log('Max retries reached, giving up')\n this.failures = 0\n // An idle socket replaces the one that gave up, so the next utterance\n // connects again with a full retry budget\n this.socket.destroy()\n this.socket = this.createSocket()\n this.options.onFailed()\n }\n}\n","import type {\n TextToSpeechStreamRequestOutputFormat,\n VoiceSettings,\n} from '@elevenlabs/elevenlabs-js/api'\n\nexport type ElevenLabsModelId =\n // Text to Dialogue WebSocket\n | 'eleven_v4_turbo'\n | 'eleven_v4'\n | 'eleven_v3_conversational'\n | 'eleven_v3'\n // Text to Speech WebSocket\n | 'eleven_flash_v2_5'\n | 'eleven_multilingual_v2'\n | 'eleven_turbo_v2_5'\n\nexport interface ElevenLabsTTSOptions {\n apiKey: string\n voiceId: string\n modelId?: ElevenLabsModelId\n language?: string\n outputFormat?: TextToSpeechStreamRequestOutputFormat\n voiceSettings?: VoiceSettings\n connectionTimeout?: number\n retryDelay?: number\n maxRetry?: number\n}\n\nexport const DEFAULT_MODEL_ID = 'eleven_turbo_v2_5'\nexport const DEFAULT_OUTPUT_FORMAT = 'pcm_16000'\n\n/** Eleven v3 and v4 only run on the Text to Dialogue WebSocket */\nexport function isDialogueModel(modelId: string): boolean {\n return modelId.startsWith('eleven_v3') || modelId.startsWith('eleven_v4')\n}\n\n// Text to Speech WebSocket\n\nexport type ElevenLabsWebSocketMessage =\n | ElevenLabsWebSocketAudioOutputMessage\n | ElevenLabsWebSocketFinalOutputMessage\n | ElevenLabsWebSocketErrorMessage\n\nexport interface ElevenLabsWebSocketAudioOutputMessage {\n audio: string\n normalizedAlignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n alignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n isFinal?: boolean | null\n}\n\nexport interface ElevenLabsWebSocketAudioOutputMessageAlignment {\n chars: string[]\n charStartTimesMs: number[]\n charDurationsMs: number[]\n}\n\nexport interface ElevenLabsWebSocketFinalOutputMessage {\n isFinal: boolean\n}\n\nexport interface ElevenLabsWebSocketErrorMessage {\n message: string\n error: string\n code: number\n}\n\n// Text to Dialogue WebSocket\n\nexport interface ElevenLabsDialogueClientMessage {\n voices?: string[]\n voice_settings?: { stability?: number }\n inputs?: ElevenLabsDialogueInput[]\n flush?: boolean\n keep_alive?: boolean\n}\n\nexport interface ElevenLabsDialogueInput {\n text: string\n voice_id: string\n new_turn?: boolean\n}\n\nexport type ElevenLabsDialogueMessage =\n | { audio: string }\n | { is_final_audio_for_turn: boolean }\n | { is_final: boolean }\n | ElevenLabsWebSocketErrorMessage\n","import { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n DEFAULT_OUTPUT_FORMAT,\n ElevenLabsDialogueClientMessage,\n ElevenLabsDialogueMessage,\n ElevenLabsTTSOptions,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-dialogue/ttd-websocket\n\n// The server closes a connection that hears nothing for 20 seconds\nconst KEEP_ALIVE_INTERVAL = 15000\n\n// Punctuation closing a sentence, with the space that follows it. An ellipsis\n// is a pause within the sentence, which v4 performs, so it does not close one.\nconst SENTENCE_END = /(?:[!?]+|(?<!\\.)\\.(?!\\.))\\s+/g\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `streaming`: the text stream is still coming\n * - `draining`: the text stream ended, the audio of its flushes is coming\n */\ntype UtteranceState = 'idle' | 'streaming' | 'draining'\n\n/**\n * The Text to Dialogue WebSocket, the only one serving Eleven v3 and v4.\n *\n * The server holds text back until it has about 40 characters and 8 words,\n * which delays a short first sentence such as \"Hi there!\". Flushing at the end\n * of every sentence starts its audio right away.\n *\n * Nothing in the protocol stops a synthesis once it is sent, so cancelling\n * drops the connection and opens a fresh one. It opens in under 100 ms, while\n * the user is still speaking.\n */\nexport class DialogueWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private unflushed = false // Whether text was sent since the last flush\n private pendingFlushes = 0 // Flushes whose audio is not complete yet\n private newTurn = false // Whether the next input starts a new utterance\n private hasSpoken = false // Whether the connection received any text\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append(\n 'output_format',\n this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT\n )\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n // Register the voice. Dialogue models only read the stability.\n const stability = this.options.voiceSettings?.stability\n const init: ElevenLabsDialogueClientMessage = {\n voices: [this.options.voiceId],\n voice_settings: stability === undefined ? undefined : { stability },\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-dialogue/stream-input?${params.toString()}`,\n tts: options,\n init,\n keepAlive: {\n interval: KEEP_ALIVE_INTERVAL,\n message: { keep_alive: true },\n },\n host,\n onOpen: () => this.onOpen(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say, for the same\n // reason as on the Text to Speech WebSocket: a stream that never carries a\n // word must not cut off the sentence still being spoken.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'streaming'\n this.newTurn = true\n return true\n }\n\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendText(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.endText()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n if (this.textBuffer.trim()) {\n this.sendText(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.endText()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n private onOpen() {\n this.hasSpoken = false\n if (this.state === 'idle') return\n // The flushes in flight were lost with the previous connection, so the\n // utterance is spoken again from its start\n const text = this.textSent\n this.textSent = ''\n this.unflushed = false\n this.pendingFlushes = 0\n if (text.trim()) {\n this.log('Sending text again')\n this.sendText(text)\n }\n if (this.state === 'draining') this.flush()\n this.checkEnded()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle' ? '' : this.textSent + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.unflushed = false\n this.pendingFlushes = 0\n }\n\n /** The text stream is over: flushes what the server holds */\n private endText() {\n this.state = 'draining'\n this.flush()\n this.checkEnded()\n }\n\n private onMessage(message: ElevenLabsDialogueMessage) {\n if ('audio' in message && message.audio) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n } else {\n this.log(`Received audio chunk (${chunk.length} bytes)`)\n this.host.audio(chunk)\n }\n }\n if (\n 'is_final_audio_for_turn' in message &&\n message.is_final_audio_for_turn\n ) {\n this.pendingFlushes = Math.max(0, this.pendingFlushes - 1)\n this.checkEnded()\n }\n // Sent before the server closes the connection, for a wrong key, an\n // unknown voice or no dialogue session left\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n /** Sends text, flushing after each sentence it completes */\n private sendText(text: string) {\n let start = 0\n for (const match of text.matchAll(SENTENCE_END)) {\n const end = match.index! + match[0].length\n this.sendInput(text.slice(start, end), true)\n start = end\n }\n if (start < text.length) {\n this.sendInput(text.slice(start), false)\n }\n }\n\n private sendInput(text: string, flush: boolean) {\n this.textSent += text\n if (!text.trim()) return\n this.send({\n inputs: [\n {\n text,\n voice_id: this.options.voiceId,\n // Resets the prosody between two utterances\n new_turn: (this.newTurn && this.hasSpoken) || undefined,\n },\n ],\n flush: flush || undefined,\n })\n this.log(`Sent text${flush ? ' and flushed' : ''}: \"${text}\"`)\n this.newTurn = false\n this.hasSpoken = true\n if (flush) {\n this.pendingFlushes++\n this.unflushed = false\n } else {\n this.unflushed = true\n }\n }\n\n /** Generates the text held by the server, if any */\n private flush() {\n if (!this.unflushed) return\n this.send({ flush: true })\n this.pendingFlushes++\n this.unflushed = false\n this.log('Flushed text')\n }\n\n /** Ends the utterance once its text is over and all its audio came */\n private checkEnded() {\n if (this.state !== 'draining') return\n if (this.unflushed || this.pendingFlushes > 0) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n // Lost when the connection just dropped, which is fine: the reconnection\n // sends the utterance again from textSent\n private send(message: ElevenLabsDialogueClientMessage) {\n this.connection.send(message)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n","import type { VoiceSettings } from '@elevenlabs/elevenlabs-js/api'\nimport { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n DEFAULT_OUTPUT_FORMAT,\n ElevenLabsTTSOptions,\n ElevenLabsWebSocketAudioOutputMessage,\n ElevenLabsWebSocketMessage,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-speech/v-1-text-to-speech-voice-id-stream-input\n\nconst WS_INACTIVITY_TIMEOUT = 180\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `speaking`: the text stream is still coming\n * - `flushed`: the text stream ended and was flushed, its audio is coming\n */\ntype UtteranceState = 'idle' | 'speaking' | 'flushed'\n\n/**\n * The Text to Speech WebSocket, for Flash, Multilingual v2 and Turbo v2.5.\n *\n * Its messages carry no utterance ID, and nothing in the protocol stops a\n * synthesis once it is sent. So cancelling drops the connection and opens a\n * fresh one, while the user is still speaking, and the audio of the cancelled\n * utterance is never heard of again.\n */\nexport class SpeechWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private heardUntil = 0 // Index in textSent where the audio received ends\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n // Build query params\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append(\n 'output_format',\n this.options.outputFormat ?? DEFAULT_OUTPUT_FORMAT\n )\n params.append('inactivity_timeout', WS_INACTIVITY_TIMEOUT.toString())\n const voiceSettings = toWireVoiceSettings(this.options.voiceSettings)\n if (voiceSettings) {\n params.append('voice_settings', JSON.stringify(voiceSettings))\n }\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-speech/${this.options.voiceId}/stream-input?${params.toString()}`,\n tts: options,\n // Initialization message with voice settings\n init: { text: ' ', voice_settings: voiceSettings },\n keepAlive: {\n interval: (WS_INACTIVITY_TIMEOUT - 1) * 1000,\n message: { text: ' ' },\n },\n host,\n onOpen: () => this.resume(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say.\n //\n // These fields track the utterance being spoken right now, and the\n // comparison between what was sent and what came back as audio is what\n // says where it is. Resetting them for a stream that never carries a word,\n // which is what an answer skipped by a tool or by onBeforeAnswer hands\n // over, loses the state of a sentence still being spoken and flushes it\n // for nothing.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'speaking'\n return true\n }\n\n // Forward text chunks coming from the caller to ElevenLabs\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendTranscript(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.flush()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n // Send last buffered text\n if (this.textBuffer.trim()) {\n this.sendTranscript(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.flush()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n /** Flushes buffered text and marks the end of the utterance */\n private flush() {\n this.state = 'flushed'\n this.connection.send({ text: ' ', flush: true })\n this.log('Flushed text')\n this.checkEnded()\n }\n\n /**\n * Sends the text of the utterance in flight again on a new connection. The\n * server lost what it was generating, so it goes on from the audio received.\n */\n private resume() {\n if (this.state === 'idle') return\n // Audio can stop inside a word, which goes again whole rather than from\n // its middle\n const wordStart = this.textSent.lastIndexOf(' ', this.heardUntil - 1) + 1\n const unheard = this.textSent.slice(wordStart)\n this.textSent = ''\n this.heardUntil = 0\n if (unheard.trim()) {\n this.log('Sending text chunks again')\n this.sendTranscript(unheard)\n }\n if (this.state === 'flushed') this.flush()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle'\n ? ''\n : this.textSent.slice(this.heardUntil) + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.heardUntil = 0\n }\n\n private onMessage(message: ElevenLabsWebSocketMessage) {\n if ('audio' in message && message.audio) {\n this.processAudioMessage(message)\n }\n if ('isFinal' in message && message.isFinal && this.state !== 'idle') {\n this.log('Audio ended')\n this.resetUtterance()\n }\n // Sent before the server closes the connection, for a wrong key or an\n // unknown voice\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n private processAudioMessage(message: ElevenLabsWebSocketAudioOutputMessage) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n return\n }\n\n // Check text of received audio chunk\n if (message.alignment) {\n const chunkText = message.alignment.chars.join('')\n\n // A connection carries one utterance at a time, since a cancel or a\n // new utterance opens another, so its audio is always wanted. The\n // alignment only tells how far the voice got, for a reconnection. It\n // follows the text sent, a few characters aside when the voice skips\n // or rewrites some, and is placed where it fits next.\n const end = findHeardEnd(this.textSent, this.heardUntil, chunkText)\n if (end === -1) {\n this.log(`Audio chunk off the text sent (\"${chunkText}\")`)\n } else {\n this.log(`Received audio chunk (\"${chunkText}\")`)\n this.heardUntil = end\n }\n }\n\n // Send to audio stream\n this.host.audio(chunk)\n\n // ElevenLabs doesn't send isFinal after a flush, so the end of the\n // utterance is when its audio covers all its text\n this.checkEnded()\n }\n\n private checkEnded() {\n if (this.state !== 'flushed') return\n if (this.textSent.slice(this.heardUntil).trim()) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n private sendTranscript(text: string) {\n // Counted as sent even when the connection just dropped, so the\n // reconnection sends it again\n this.textSent += text\n if (!this.connection.send({ text, try_trigger_generation: true })) return\n this.log(`Sent transcript: \"${text}\"`)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n\n/**\n * Where `heard` ends in `text`, at the first place from `from` where the text\n * goes on with it, or -1 when it does nowhere\n */\nfunction findHeardEnd(text: string, from: number, heard: string): number {\n for (let start = from; start < text.length; start++) {\n const end = endOfPrefix(text, start, heard)\n if (end !== -1) return end\n }\n return -1\n}\n\n/**\n * Where `heard` ends in `text`, read from `from` and compared without\n * whitespace, or -1 when the text does not go on with it there.\n */\nfunction endOfPrefix(text: string, from: number, heard: string): number {\n let index = from\n for (let i = 0; i < heard.length; i++) {\n if (isSpace(heard[i])) continue\n while (index < text.length && isSpace(text[index])) index++\n if (text[index] !== heard[i]) return -1\n index++\n }\n return index\n}\n\nfunction isSpace(char: string) {\n return char === ' ' || char === '\\t' || char === '\\n' || char === '\\r'\n}\n\n/**\n * Voice settings as the WebSocket reads them.\n *\n * The SDK types them in camel case and the WebSocket reads snake case, so\n * similarityBoost and useSpeakerBoost would be ignored if sent as they are.\n */\nfunction toWireVoiceSettings(settings?: VoiceSettings) {\n if (!settings) return undefined\n const { similarityBoost, useSpeakerBoost, ...rest } = settings\n return {\n similarity_boost: similarityBoost,\n use_speaker_boost: useSpeakerBoost,\n ...rest,\n }\n}\n"],"mappings":";AAAA,SAAS,WAAW;;;ACApB,SAAS,0BAA0B;AAInC,IAAM,oBAAoB;AAuCnB,IAAM,uBAAN,MAA2B;AAAA,EAOhC,YAA6B,SAAsC;AAAtC;AAH7B;AAAA;AAAA,SAAQ,SAAS;AACjB,SAAQ,WAAW;AAGjB,SAAK,SAAS,KAAK,aAAa;AAAA,EAClC;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,QAA0B;AACxB,QAAI,KAAK,OAAO,WAAW,OAAQ,QAAO,QAAQ,QAAQ,KAAK;AAC/D,WAAO,KAAK,OAAO,MAAM;AAAA,EAC3B;AAAA,EAEA,KAAK,SAA0B;AAC7B,WAAO,KAAK,OAAO,KAAK,KAAK,UAAU,OAAO,CAAC;AAAA,EACjD;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA,EAEA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,SAAS;AACd,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,eAAe;AACrB,UAAM,EAAE,KAAK,WAAW,KAAK,IAAI,KAAK;AACtC,UAAM,SAAS,IAAI,mBAAmB;AAAA,MACpC,KAAK,KAAK,QAAQ;AAAA,MAClB,eAAe,EAAE,SAAS,EAAE,cAAc,IAAI,OAAO,EAAE;AAAA,MACvD,mBAAmB,IAAI;AAAA,MACvB,YAAY,IAAI;AAAA,MAChB,UAAU,IAAI;AAAA,MACd,WAAW;AAAA,QACT,OAAO,CAAC,OAAO,GAAG,KAAK,KAAK,UAAU,KAAK,QAAQ,IAAI,CAAC;AAAA,MAC1D;AAAA,MACA,WAAW;AAAA,QACT,UAAU,UAAU;AAAA,QACpB,SAAS,MAAM;AACb,eAAK,MAAM;AACX,iBAAO,KAAK,UAAU,UAAU,OAAO;AAAA,QACzC;AAAA,MACF;AAAA,MACA,KAAK,IAAI,YAAY,KAAK,IAAI,GAAG,OAAO;AAAA,IAC1C,CAAC;AAED,WAAO,GAAG,QAAQ,CAAC,EAAE,YAAY,MAAM;AACrC,WAAK,SAAS;AACd,WAAK,QAAQ,OAAO,WAAW;AAAA,IACjC,CAAC;AAED,WAAO,GAAG,WAAW,CAAC,SAAS;AAC7B,UAAI;AACF,aAAK,QAAQ,UAAU,KAAK,MAAM,KAAK,SAAS,CAAC,CAAC;AAAA,MACpD,SAAS,OAAY;AACnB,aAAK,IAAI,aAAa,QAAQ,MAAM,UAAU,KAAK;AACnD,aAAK,IAAI,4BAA4B,KAAK,SAAS,CAAC;AAAA,MACtD;AAAA,IACF,CAAC;AAED,WAAO,GAAG,SAAS,MAAM;AACvB,UAAI,KAAK,OAAQ;AACjB,WAAK;AACL,UAAI,KAAK,YAAY,IAAI,YAAY,mBAAoB,MAAK,OAAO;AAAA,IACvE,CAAC;AAED,WAAO,GAAG,UAAU,MAAM,KAAK,OAAO,CAAC;AACvC,WAAO;AAAA,EACT;AAAA,EAEQ,SAAS;AACf,SAAK,QAAQ,KAAK,IAAI,gCAAgC;AACtD,SAAK,WAAW;AAGhB,SAAK,OAAO,QAAQ;AACpB,SAAK,SAAS,KAAK,aAAa;AAChC,SAAK,QAAQ,SAAS;AAAA,EACxB;AACF;;;AClHO,IAAM,mBAAmB;AACzB,IAAM,wBAAwB;AAG9B,SAAS,gBAAgB,SAA0B;AACxD,SAAO,QAAQ,WAAW,WAAW,KAAK,QAAQ,WAAW,WAAW;AAC1E;;;ACjBA,IAAM,sBAAsB;AAI5B,IAAM,eAAe;AAsBd,IAAM,oBAAN,MAAuD;AAAA;AAAA,EAgB5D,YACmB,SACA,MACjB;AAFiB;AACA;AAhBnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,YAAY;AACpB;AAAA,SAAQ,iBAAiB;AACzB;AAAA,SAAQ,UAAU;AAClB;AAAA,SAAQ,YAAY;AAMlB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO;AAAA,MACL;AAAA,MACA,KAAK,QAAQ,gBAAgB;AAAA,IAC/B;AACA,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAGA,UAAM,YAAY,KAAK,QAAQ,eAAe;AAC9C,UAAM,OAAwC;AAAA,MAC5C,QAAQ,CAAC,KAAK,QAAQ,OAAO;AAAA,MAC7B,gBAAgB,cAAc,SAAY,SAAY,EAAE,UAAU;AAAA,IACpE;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,4DAA4D,OAAO,SAAS,CAAC;AAAA,MAClF,KAAK;AAAA,MACL;AAAA,MACA,WAAW;AAAA,QACT,UAAU;AAAA,QACV,SAAS,EAAE,YAAY,KAAK;AAAA,MAC9B;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAKd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,WAAK,UAAU;AACf,aAAO;AAAA,IACT;AAEA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,SAAS,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AACtD,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAClC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,SAAS,KAAK,aAAa,GAAG;AAAA,MACrC;AACA,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEQ,SAAS;AACf,SAAK,YAAY;AACjB,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,OAAO,KAAK;AAClB,SAAK,WAAW;AAChB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AACtB,QAAI,KAAK,KAAK,GAAG;AACf,WAAK,IAAI,oBAAoB;AAC7B,WAAK,SAAS,IAAI;AAAA,IACpB;AACA,QAAI,KAAK,UAAU,WAAY,MAAK,MAAM;AAC1C,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SAAS,KAAK,KAAK,WAAW,KAAK;AACpD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AAAA,EACxB;AAAA;AAAA,EAGQ,UAAU;AAChB,SAAK,QAAQ;AACb,SAAK,MAAM;AACX,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,UAAU,SAAoC;AACpD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,YAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,WAAK,WAAW,MAAM;AACtB,UAAI,KAAK,UAAU,QAAQ;AACzB,aAAK,IAAI,6CAA6C;AAAA,MACxD,OAAO;AACL,aAAK,IAAI,yBAAyB,MAAM,MAAM,SAAS;AACvD,aAAK,KAAK,MAAM,KAAK;AAAA,MACvB;AAAA,IACF;AACA,QACE,6BAA6B,WAC7B,QAAQ,yBACR;AACA,WAAK,iBAAiB,KAAK,IAAI,GAAG,KAAK,iBAAiB,CAAC;AACzD,WAAK,WAAW;AAAA,IAClB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA;AAAA,EAGQ,SAAS,MAAc;AAC7B,QAAI,QAAQ;AACZ,eAAW,SAAS,KAAK,SAAS,YAAY,GAAG;AAC/C,YAAM,MAAM,MAAM,QAAS,MAAM,CAAC,EAAE;AACpC,WAAK,UAAU,KAAK,MAAM,OAAO,GAAG,GAAG,IAAI;AAC3C,cAAQ;AAAA,IACV;AACA,QAAI,QAAQ,KAAK,QAAQ;AACvB,WAAK,UAAU,KAAK,MAAM,KAAK,GAAG,KAAK;AAAA,IACzC;AAAA,EACF;AAAA,EAEQ,UAAU,MAAc,OAAgB;AAC9C,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,KAAK,EAAG;AAClB,SAAK,KAAK;AAAA,MACR,QAAQ;AAAA,QACN;AAAA,UACE;AAAA,UACA,UAAU,KAAK,QAAQ;AAAA;AAAA,UAEvB,UAAW,KAAK,WAAW,KAAK,aAAc;AAAA,QAChD;AAAA,MACF;AAAA,MACA,OAAO,SAAS;AAAA,IAClB,CAAC;AACD,SAAK,IAAI,YAAY,QAAQ,iBAAiB,EAAE,MAAM,IAAI,GAAG;AAC7D,SAAK,UAAU;AACf,SAAK,YAAY;AACjB,QAAI,OAAO;AACT,WAAK;AACL,WAAK,YAAY;AAAA,IACnB,OAAO;AACL,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGQ,QAAQ;AACd,QAAI,CAAC,KAAK,UAAW;AACrB,SAAK,KAAK,EAAE,OAAO,KAAK,CAAC;AACzB,SAAK;AACL,SAAK,YAAY;AACjB,SAAK,IAAI,cAAc;AAAA,EACzB;AAAA;AAAA,EAGQ,aAAa;AACnB,QAAI,KAAK,UAAU,WAAY;AAC/B,QAAI,KAAK,aAAa,KAAK,iBAAiB,EAAG;AAC/C,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA;AAAA,EAIQ,KAAK,SAA0C;AACrD,SAAK,WAAW,KAAK,OAAO;AAAA,EAC9B;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;;;AC3RA,IAAM,wBAAwB;AAmBvB,IAAM,kBAAN,MAAqD;AAAA;AAAA,EAa1D,YACmB,SACA,MACjB;AAFiB;AACA;AAbnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,aAAa;AAOnB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO;AAAA,MACL;AAAA,MACA,KAAK,QAAQ,gBAAgB;AAAA,IAC/B;AACA,WAAO,OAAO,sBAAsB,sBAAsB,SAAS,CAAC;AACpE,UAAM,gBAAgB,oBAAoB,KAAK,QAAQ,aAAa;AACpE,QAAI,eAAe;AACjB,aAAO,OAAO,kBAAkB,KAAK,UAAU,aAAa,CAAC;AAAA,IAC/D;AACA,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,6CAA6C,KAAK,QAAQ,OAAO,iBAAiB,OAAO,SAAS,CAAC;AAAA,MACxG,KAAK;AAAA;AAAA,MAEL,MAAM,EAAE,MAAM,KAAK,gBAAgB,cAAc;AAAA,MACjD,WAAW;AAAA,QACT,WAAW,wBAAwB,KAAK;AAAA,QACxC,SAAS,EAAE,MAAM,IAAI;AAAA,MACvB;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAUd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,aAAO;AAAA,IACT;AAGA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,eAAe,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AAC5D,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAElC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,eAAe,KAAK,aAAa,GAAG;AAAA,MAC3C;AACA,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA;AAAA,EAGQ,QAAQ;AACd,SAAK,QAAQ;AACb,SAAK,WAAW,KAAK,EAAE,MAAM,KAAK,OAAO,KAAK,CAAC;AAC/C,SAAK,IAAI,cAAc;AACvB,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMQ,SAAS;AACf,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,YAAY,KAAK,SAAS,YAAY,KAAK,KAAK,aAAa,CAAC,IAAI;AACxE,UAAM,UAAU,KAAK,SAAS,MAAM,SAAS;AAC7C,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,QAAI,QAAQ,KAAK,GAAG;AAClB,WAAK,IAAI,2BAA2B;AACpC,WAAK,eAAe,OAAO;AAAA,IAC7B;AACA,QAAI,KAAK,UAAU,UAAW,MAAK,MAAM;AAAA,EAC3C;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SACX,KACA,KAAK,SAAS,MAAM,KAAK,UAAU,IAAI,KAAK;AAClD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,aAAa;AAAA,EACpB;AAAA,EAEQ,UAAU,SAAqC;AACrD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,WAAK,oBAAoB,OAAO;AAAA,IAClC;AACA,QAAI,aAAa,WAAW,QAAQ,WAAW,KAAK,UAAU,QAAQ;AACpE,WAAK,IAAI,aAAa;AACtB,WAAK,eAAe;AAAA,IACtB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA,EAEQ,oBAAoB,SAAgD;AAC1E,UAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,SAAK,WAAW,MAAM;AACtB,QAAI,KAAK,UAAU,QAAQ;AACzB,WAAK,IAAI,6CAA6C;AACtD;AAAA,IACF;AAGA,QAAI,QAAQ,WAAW;AACrB,YAAM,YAAY,QAAQ,UAAU,MAAM,KAAK,EAAE;AAOjD,YAAM,MAAM,aAAa,KAAK,UAAU,KAAK,YAAY,SAAS;AAClE,UAAI,QAAQ,IAAI;AACd,aAAK,IAAI,mCAAmC,SAAS,IAAI;AAAA,MAC3D,OAAO;AACL,aAAK,IAAI,0BAA0B,SAAS,IAAI;AAChD,aAAK,aAAa;AAAA,MACpB;AAAA,IACF;AAGA,SAAK,KAAK,MAAM,KAAK;AAIrB,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,aAAa;AACnB,QAAI,KAAK,UAAU,UAAW;AAC9B,QAAI,KAAK,SAAS,MAAM,KAAK,UAAU,EAAE,KAAK,EAAG;AACjD,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA,EAEQ,eAAe,MAAc;AAGnC,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,WAAW,KAAK,EAAE,MAAM,wBAAwB,KAAK,CAAC,EAAG;AACnE,SAAK,IAAI,qBAAqB,IAAI,GAAG;AAAA,EACvC;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;AAMA,SAAS,aAAa,MAAc,MAAc,OAAuB;AACvE,WAAS,QAAQ,MAAM,QAAQ,KAAK,QAAQ,SAAS;AACnD,UAAM,MAAM,YAAY,MAAM,OAAO,KAAK;AAC1C,QAAI,QAAQ,GAAI,QAAO;AAAA,EACzB;AACA,SAAO;AACT;AAMA,SAAS,YAAY,MAAc,MAAc,OAAuB;AACtE,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,QAAI,QAAQ,MAAM,CAAC,CAAC,EAAG;AACvB,WAAO,QAAQ,KAAK,UAAU,QAAQ,KAAK,KAAK,CAAC,EAAG;AACpD,QAAI,KAAK,KAAK,MAAM,MAAM,CAAC,EAAG,QAAO;AACrC;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,QAAQ,MAAc;AAC7B,SAAO,SAAS,OAAO,SAAS,OAAQ,SAAS,QAAQ,SAAS;AACpE;AAQA,SAAS,oBAAoB,UAA0B;AACrD,MAAI,CAAC,SAAU,QAAO;AACtB,QAAM,EAAE,iBAAiB,iBAAiB,GAAG,KAAK,IAAI;AACtD,SAAO;AAAA,IACL,kBAAkB;AAAA,IAClB,mBAAmB;AAAA,IACnB,GAAG;AAAA,EACL;AACF;;;AJnTO,IAAM,gBAAN,cAA4B,IAAI;AAAA,EAGrC,YAAY,SAA+B;AACzC,UAAM;AACN,UAAM,OAAO;AAAA,MACX,OAAO,CAAC,UAAkB,KAAK,KAAK,SAAS,KAAK;AAAA,MAClD,QAAQ,CAAC,UAAoB,KAAK,KAAK,UAAU,KAAK;AAAA,MACtD,KAAK,IAAI,YAAmB,KAAK,IAAI,GAAG,OAAO;AAAA,IACjD;AACA,SAAK,YAAY,gBAAgB,QAAQ,WAAW,gBAAgB,IAChE,IAAI,kBAAkB,SAAS,IAAI,IACnC,IAAI,gBAAgB,SAAS,IAAI;AAAA,EACvC;AAAA,EAEA,MAAM,YAAsB;AAC1B,SAAK,UAAU,MAAM,UAAU;AAAA,EACjC;AAAA,EAEA,SAAS;AACP,SAAK,UAAU,OAAO;AAAA,EACxB;AAAA,EAEA,UAAU;AACR,SAAK,UAAU,QAAQ;AACvB,UAAM,QAAQ;AAAA,EAChB;AACF;","names":[]}
|
|
1
|
+
{"version":3,"sources":["../src/ElevenLabsTTS.ts","../src/ElevenLabsWebSocket.ts","../src/types.ts","../src/DialogueWebSocket.ts","../src/SpeechWebSocket.ts"],"sourcesContent":["import {\n cachedModels,\n cachedVoices,\n MicdropModel,\n MicdropVoice,\n TTS,\n voiceGender,\n voicesForLanguage,\n} from '@micdrop/server'\nimport { Readable } from 'stream'\nimport { DialogueWebSocket } from './DialogueWebSocket'\nimport { ElevenLabsWebSocket } from './ElevenLabsWebSocket'\nimport { SpeechWebSocket } from './SpeechWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsTTSOptions,\n getSampleRate,\n isDialogueModel,\n} from './types'\n\n/** A voice as the ElevenLabs API lists it, the fields read here */\ninterface ElevenLabsVoice {\n voice_id: string\n name: string\n category?: string\n description?: string | null\n preview_url?: string | null\n labels?: Record<string, string | undefined>\n verified_languages?: Array<{ language?: string; locale?: string | null }>\n}\n\n// Voices made in the account rather than offered to everyone\nconst CUSTOM_CATEGORIES = ['cloned', 'generated', 'professional']\n\n/**\n * ElevenLabs text to speech, streamed over the WebSocket the model runs on.\n *\n * Eleven v3 and v4 only answer on the Text to Dialogue WebSocket, the other\n * models on the Text to Speech WebSocket. The two speak different protocols,\n * so the model picks the one this class drives.\n */\nexport class ElevenLabsTTS extends TTS {\n /**\n * Lists the models that speak, with the languages each one reads.\n *\n * The list is fetched once and shared by every call of the process for a\n * day. `refresh` fetches it again.\n */\n static listModels(options: {\n apiKey: string\n refresh?: boolean\n }): Promise<MicdropModel[]> {\n return cachedModels(\n ['elevenlabs'],\n options.apiKey,\n async () => {\n const response = await fetch('https://api.elevenlabs.io/v1/models', {\n headers: { 'xi-api-key': options.apiKey },\n })\n if (!response.ok) {\n throw new Error(`ElevenLabs answered ${response.status}`)\n }\n const models = (await response.json()) as Array<{\n model_id: string\n name?: string\n description?: string\n can_do_text_to_speech?: boolean\n languages?: Array<{ language_id: string }>\n }>\n return models\n .filter((model) => model.can_do_text_to_speech)\n .map((model) => ({\n id: model.model_id,\n name: model.name,\n description: model.description,\n languages: model.languages?.map(({ language_id }) => language_id),\n }))\n },\n options.refresh\n )\n }\n\n /**\n * Lists the voices the account can speak with: the premade ones, and the\n * ones cloned, designed or added from the library.\n *\n * `language`, such as `fr` or `fr-FR`, keeps the voices that speak it.\n *\n * The list is fetched once and shared by every call of the process for a\n * day. `refresh` fetches it again, for a voice added a moment ago.\n */\n static listVoices(options: {\n apiKey: string\n language?: string\n refresh?: boolean\n }): Promise<MicdropVoice[]> {\n return cachedVoices(\n ['elevenlabs'],\n options.apiKey,\n () => fetchVoices(options.apiKey),\n options.refresh\n ).then((voices) => voicesForLanguage(voices, options.language))\n }\n\n private readonly webSocket: ElevenLabsWebSocket\n private readonly rate: number\n\n constructor(options: ElevenLabsTTSOptions) {\n super()\n this.rate = getSampleRate(options)\n const host = {\n audio: (chunk: Buffer) => this.emit('Audio', chunk),\n failed: (texts: string[]) => this.emit('Failed', texts),\n log: (...message: any[]) => this.log(...message),\n }\n this.webSocket = isDialogueModel(options.modelId ?? DEFAULT_MODEL_ID)\n ? new DialogueWebSocket(options, host)\n : new SpeechWebSocket(options, host)\n }\n\n get sampleRate() {\n return this.rate\n }\n\n speak(textStream: Readable) {\n this.webSocket.speak(textStream)\n }\n\n cancel() {\n this.webSocket.cancel()\n }\n\n destroy() {\n this.webSocket.destroy()\n super.destroy()\n }\n}\n\n/** Reads every page of the voices of the account */\nasync function fetchVoices(apiKey: string): Promise<MicdropVoice[]> {\n const voices: MicdropVoice[] = []\n let pageToken: string | undefined\n do {\n const params = new URLSearchParams({ page_size: '100' })\n if (pageToken) params.set('next_page_token', pageToken)\n const response = await fetch(\n `https://api.elevenlabs.io/v2/voices?${params}`,\n { headers: { 'xi-api-key': apiKey } }\n )\n if (!response.ok) throw new Error(`ElevenLabs answered ${response.status}`)\n const page = (await response.json()) as {\n voices: ElevenLabsVoice[]\n has_more?: boolean\n next_page_token?: string | null\n }\n voices.push(...page.voices.map(toMicdropVoice))\n pageToken = page.has_more ? (page.next_page_token ?? undefined) : undefined\n } while (pageToken)\n return voices\n}\n\nfunction toMicdropVoice(voice: ElevenLabsVoice): MicdropVoice {\n const verified = (voice.verified_languages ?? [])\n .map(({ locale, language }) => locale || language)\n .filter((code): code is string => !!code)\n const label = voice.labels?.language\n const languages = verified.length ? verified : label ? [label] : undefined\n return {\n id: voice.voice_id,\n name: voice.name,\n languages: languages && [...new Set(languages)],\n gender: voiceGender(voice.labels?.gender),\n description: voice.description || undefined,\n previewUrl: voice.preview_url || undefined,\n custom: CUSTOM_CATEGORIES.includes(voice.category ?? ''),\n }\n}\n","import { ReconnectingSocket } from '@micdrop/server'\nimport type { Readable } from 'stream'\nimport type { ElevenLabsTTSOptions } from './types'\n\nconst DEFAULT_MAX_RETRY = 3\n\n/** One of the two ElevenLabs WebSockets, driven by ElevenLabsTTS */\nexport interface ElevenLabsWebSocket {\n speak(textStream: Readable): void\n cancel(): void\n destroy(): void\n}\n\n/** What a WebSocket hands back to the ElevenLabsTTS driving it */\nexport interface ElevenLabsWebSocketHost {\n audio(chunk: Buffer): void\n failed(texts: string[]): void\n log(...message: any[]): void\n}\n\nexport interface ElevenLabsConnectionOptions {\n url: string\n tts: ElevenLabsTTSOptions\n /** First message of every connection, which sets the session up */\n init: object\n keepAlive: { interval: number; message: object }\n host: ElevenLabsWebSocketHost\n /** The connection opened, after a drop when `reconnected` is true */\n onOpen(reconnected: boolean): void\n onMessage(message: any): void\n /** Every retry failed, and the connection waits for `connect()` */\n onFailed(): void\n}\n\n/**\n * The connection of one of the two WebSockets, reconnecting on its own.\n *\n * ElevenLabs opens a connection before it checks the key and the voice, then\n * closes it with an error. Such a connection opens fine, which would start the\n * retry count of ReconnectingSocket over each time and reconnect forever. So a\n * connection closed before it proved usable counts as a failed one, and the\n * connection gives up once they exceed `maxRetry` in a row.\n */\nexport class ElevenLabsConnection {\n private socket: ReconnectingSocket\n // Whether the current connection carried audio or lived long enough for a\n // keep-alive, which means the server accepted the key and the voice\n private proven = false\n private failures = 0\n\n constructor(private readonly options: ElevenLabsConnectionOptions) {\n this.socket = this.createSocket()\n }\n\n /** Opens the connection, again after giving up, and does nothing otherwise */\n connect() {\n this.socket.connect()\n }\n\n /**\n * Resolves to false when the connection gives up or is destroyed meanwhile,\n * and right away once it gave up, until the next `connect()`: the stream\n * that was speaking then has nothing left to wait for.\n */\n ready(): Promise<boolean> {\n if (this.socket.status === 'idle') return Promise.resolve(false)\n return this.socket.ready()\n }\n\n send(message: object): boolean {\n return this.socket.send(JSON.stringify(message))\n }\n\n /** Drops the connection, with the synthesis running on it, for a new one */\n restart() {\n this.socket.restart()\n }\n\n destroy() {\n this.socket.destroy()\n }\n\n /** The current connection works, so its next drop is not a failure */\n prove() {\n this.proven = true\n this.failures = 0\n }\n\n private createSocket() {\n const { tts, keepAlive, host } = this.options\n const socket = new ReconnectingSocket({\n url: this.options.url,\n clientOptions: { headers: { 'xi-api-key': tts.apiKey } },\n connectionTimeout: tts.connectionTimeout,\n retryDelay: tts.retryDelay,\n maxRetry: tts.maxRetry,\n handshake: {\n start: (ws) => ws.send(JSON.stringify(this.options.init)),\n },\n keepAlive: {\n interval: keepAlive.interval,\n message: () => {\n this.prove()\n return JSON.stringify(keepAlive.message)\n },\n },\n log: (...message) => host.log(...message),\n })\n\n socket.on('Open', ({ reconnected }) => {\n this.proven = false\n this.options.onOpen(reconnected)\n })\n\n socket.on('Message', (data) => {\n try {\n this.options.onMessage(JSON.parse(data.toString()))\n } catch (error: any) {\n host.log('message' in error ? error.message : error)\n host.log('Event data during error:', data.toString())\n }\n })\n\n socket.on('Close', () => {\n if (this.proven) return\n this.failures++\n if (this.failures > (tts.maxRetry ?? DEFAULT_MAX_RETRY)) this.giveUp()\n })\n\n socket.on('Failed', () => this.giveUp())\n return socket\n }\n\n private giveUp() {\n this.options.host.log('Max retries reached, giving up')\n this.failures = 0\n // An idle socket replaces the one that gave up, so the next utterance\n // connects again with a full retry budget\n this.socket.destroy()\n this.socket = this.createSocket()\n this.options.onFailed()\n }\n}\n","import type {\n TextToSpeechStreamRequestOutputFormat,\n VoiceSettings,\n} from '@elevenlabs/elevenlabs-js/api'\n\nexport type ElevenLabsModelId =\n // Text to Dialogue WebSocket\n | 'eleven_v4_turbo'\n | 'eleven_v4'\n | 'eleven_v3_conversational'\n | 'eleven_v3'\n // Text to Speech WebSocket\n | 'eleven_flash_v2_5'\n | 'eleven_multilingual_v2'\n | 'eleven_turbo_v2_5'\n\n/** Rates ElevenLabs streams raw PCM at. 44100 needs a Pro plan or above. */\nexport type ElevenLabsSampleRate =\n 8000 | 16000 | 22050 | 24000 | 32000 | 44100 | 48000\n\nexport interface ElevenLabsTTSOptions {\n apiKey: string\n voiceId: string\n modelId?: ElevenLabsModelId\n language?: string\n /**\n * Rate of the audio, in Hz, 48000 by default, the highest one offered. It\n * goes to the client as is, so the voice is never resampled on the server.\n */\n sampleRate?: ElevenLabsSampleRate\n /**\n * @deprecated Use `sampleRate`. The client only plays raw PCM, so only the\n * `pcm_*` formats work.\n */\n outputFormat?: TextToSpeechStreamRequestOutputFormat\n voiceSettings?: VoiceSettings\n connectionTimeout?: number\n retryDelay?: number\n maxRetry?: number\n}\n\nexport const DEFAULT_MODEL_ID = 'eleven_turbo_v2_5'\nexport const DEFAULT_SAMPLE_RATE: ElevenLabsSampleRate = 48000\n/** @deprecated Use `DEFAULT_SAMPLE_RATE` */\nexport const DEFAULT_OUTPUT_FORMAT = 'pcm_48000'\n\n/** The `output_format` asked of ElevenLabs */\nexport function getOutputFormat(\n options: ElevenLabsTTSOptions\n): TextToSpeechStreamRequestOutputFormat {\n if (options.sampleRate) return `pcm_${options.sampleRate}`\n return options.outputFormat ?? `pcm_${DEFAULT_SAMPLE_RATE}`\n}\n\n/** Rate of the audio ElevenLabs sends, read from its `output_format` */\nexport function getSampleRate(options: ElevenLabsTTSOptions): number {\n const match = /^pcm_(\\d+)$/.exec(getOutputFormat(options))\n return match ? Number(match[1]) : DEFAULT_SAMPLE_RATE\n}\n\n/** Eleven v3 and v4 only run on the Text to Dialogue WebSocket */\nexport function isDialogueModel(modelId: string): boolean {\n return modelId.startsWith('eleven_v3') || modelId.startsWith('eleven_v4')\n}\n\n// Text to Speech WebSocket\n\nexport type ElevenLabsWebSocketMessage =\n | ElevenLabsWebSocketAudioOutputMessage\n | ElevenLabsWebSocketFinalOutputMessage\n | ElevenLabsWebSocketErrorMessage\n\nexport interface ElevenLabsWebSocketAudioOutputMessage {\n audio: string\n normalizedAlignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n alignment: ElevenLabsWebSocketAudioOutputMessageAlignment | null\n isFinal?: boolean | null\n}\n\nexport interface ElevenLabsWebSocketAudioOutputMessageAlignment {\n chars: string[]\n charStartTimesMs: number[]\n charDurationsMs: number[]\n}\n\nexport interface ElevenLabsWebSocketFinalOutputMessage {\n isFinal: boolean\n}\n\nexport interface ElevenLabsWebSocketErrorMessage {\n message: string\n error: string\n code: number\n}\n\n// Text to Dialogue WebSocket\n\nexport interface ElevenLabsDialogueClientMessage {\n voices?: string[]\n voice_settings?: { stability?: number }\n inputs?: ElevenLabsDialogueInput[]\n flush?: boolean\n keep_alive?: boolean\n}\n\nexport interface ElevenLabsDialogueInput {\n text: string\n voice_id: string\n new_turn?: boolean\n}\n\nexport type ElevenLabsDialogueMessage =\n | { audio: string }\n | { is_final_audio_for_turn: boolean }\n | { is_final: boolean }\n | ElevenLabsWebSocketErrorMessage\n","import { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsDialogueClientMessage,\n ElevenLabsDialogueMessage,\n ElevenLabsTTSOptions,\n getOutputFormat,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-dialogue/ttd-websocket\n\n// The server closes a connection that hears nothing for 20 seconds\nconst KEEP_ALIVE_INTERVAL = 15000\n\n// Punctuation closing a sentence, with the space that follows it. An ellipsis\n// is a pause within the sentence, which v4 performs, so it does not close one.\nconst SENTENCE_END = /(?:[!?]+|(?<!\\.)\\.(?!\\.))\\s+/g\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `streaming`: the text stream is still coming\n * - `draining`: the text stream ended, the audio of its flushes is coming\n */\ntype UtteranceState = 'idle' | 'streaming' | 'draining'\n\n/**\n * The Text to Dialogue WebSocket, the only one serving Eleven v3 and v4.\n *\n * The server holds text back until it has about 40 characters and 8 words,\n * which delays a short first sentence such as \"Hi there!\". Flushing at the end\n * of every sentence starts its audio right away.\n *\n * Nothing in the protocol stops a synthesis once it is sent, so cancelling\n * drops the connection and opens a fresh one. It opens in under 100 ms, while\n * the user is still speaking.\n */\nexport class DialogueWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private unflushed = false // Whether text was sent since the last flush\n private pendingFlushes = 0 // Flushes whose audio is not complete yet\n private newTurn = false // Whether the next input starts a new utterance\n private hasSpoken = false // Whether the connection received any text\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append('output_format', getOutputFormat(this.options))\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n // Register the voice. Dialogue models only read the stability.\n const stability = this.options.voiceSettings?.stability\n const init: ElevenLabsDialogueClientMessage = {\n voices: [this.options.voiceId],\n voice_settings: stability === undefined ? undefined : { stability },\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-dialogue/stream-input?${params.toString()}`,\n tts: options,\n init,\n keepAlive: {\n interval: KEEP_ALIVE_INTERVAL,\n message: { keep_alive: true },\n },\n host,\n onOpen: () => this.onOpen(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say, for the same\n // reason as on the Text to Speech WebSocket: a stream that never carries a\n // word must not cut off the sentence still being spoken.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'streaming'\n this.newTurn = true\n return true\n }\n\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendText(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.endText()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n if (this.textBuffer.trim()) {\n this.sendText(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.endText()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n private onOpen() {\n this.hasSpoken = false\n if (this.state === 'idle') return\n // The flushes in flight were lost with the previous connection, so the\n // utterance is spoken again from its start\n const text = this.textSent\n this.textSent = ''\n this.unflushed = false\n this.pendingFlushes = 0\n if (text.trim()) {\n this.log('Sending text again')\n this.sendText(text)\n }\n if (this.state === 'draining') this.flush()\n this.checkEnded()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle' ? '' : this.textSent + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.unflushed = false\n this.pendingFlushes = 0\n }\n\n /** The text stream is over: flushes what the server holds */\n private endText() {\n this.state = 'draining'\n this.flush()\n this.checkEnded()\n }\n\n private onMessage(message: ElevenLabsDialogueMessage) {\n if ('audio' in message && message.audio) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n } else {\n this.log(`Received audio chunk (${chunk.length} bytes)`)\n this.host.audio(chunk)\n }\n }\n if (\n 'is_final_audio_for_turn' in message &&\n message.is_final_audio_for_turn\n ) {\n this.pendingFlushes = Math.max(0, this.pendingFlushes - 1)\n this.checkEnded()\n }\n // Sent before the server closes the connection, for a wrong key, an\n // unknown voice or no dialogue session left\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n /** Sends text, flushing after each sentence it completes */\n private sendText(text: string) {\n let start = 0\n for (const match of text.matchAll(SENTENCE_END)) {\n const end = match.index! + match[0].length\n this.sendInput(text.slice(start, end), true)\n start = end\n }\n if (start < text.length) {\n this.sendInput(text.slice(start), false)\n }\n }\n\n private sendInput(text: string, flush: boolean) {\n this.textSent += text\n if (!text.trim()) return\n this.send({\n inputs: [\n {\n text,\n voice_id: this.options.voiceId,\n // Resets the prosody between two utterances\n new_turn: (this.newTurn && this.hasSpoken) || undefined,\n },\n ],\n flush: flush || undefined,\n })\n this.log(`Sent text${flush ? ' and flushed' : ''}: \"${text}\"`)\n this.newTurn = false\n this.hasSpoken = true\n if (flush) {\n this.pendingFlushes++\n this.unflushed = false\n } else {\n this.unflushed = true\n }\n }\n\n /** Generates the text held by the server, if any */\n private flush() {\n if (!this.unflushed) return\n this.send({ flush: true })\n this.pendingFlushes++\n this.unflushed = false\n this.log('Flushed text')\n }\n\n /** Ends the utterance once its text is over and all its audio came */\n private checkEnded() {\n if (this.state !== 'draining') return\n if (this.unflushed || this.pendingFlushes > 0) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n // Lost when the connection just dropped, which is fine: the reconnection\n // sends the utterance again from textSent\n private send(message: ElevenLabsDialogueClientMessage) {\n this.connection.send(message)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n","import type { VoiceSettings } from '@elevenlabs/elevenlabs-js/api'\nimport { Readable } from 'stream'\nimport {\n ElevenLabsConnection,\n ElevenLabsWebSocket,\n ElevenLabsWebSocketHost,\n} from './ElevenLabsWebSocket'\nimport {\n DEFAULT_MODEL_ID,\n ElevenLabsTTSOptions,\n getOutputFormat,\n ElevenLabsWebSocketAudioOutputMessage,\n ElevenLabsWebSocketMessage,\n} from './types'\n\n// API Reference: https://elevenlabs.io/docs/api-reference/text-to-speech/v-1-text-to-speech-voice-id-stream-input\n\nconst WS_INACTIVITY_TIMEOUT = 180\n\n/**\n * Where the utterance stands.\n *\n * - `idle`: nothing is being spoken\n * - `speaking`: the text stream is still coming\n * - `flushed`: the text stream ended and was flushed, its audio is coming\n */\ntype UtteranceState = 'idle' | 'speaking' | 'flushed'\n\n/**\n * The Text to Speech WebSocket, for Flash, Multilingual v2 and Turbo v2.5.\n *\n * Its messages carry no utterance ID, and nothing in the protocol stops a\n * synthesis once it is sent. So cancelling drops the connection and opens a\n * fresh one, while the user is still speaking, and the audio of the cancelled\n * utterance is never heard of again.\n */\nexport class SpeechWebSocket implements ElevenLabsWebSocket {\n private readonly connection: ElevenLabsConnection\n private state: UtteranceState = 'idle'\n // Bumped by every speak() and every cancel(), so a stream that starts late\n // can tell whether it is still the one that should be heard.\n private generation = 0\n // Generation of the stream being spoken. A speak() that never carries a\n // word leaves it in place, so the sentence it sends goes on to its end.\n private speaking = 0\n private textSent = '' // Text of the utterance sent to ElevenLabs\n private textBuffer = '' // Text not sent yet, held until its word is complete\n private heardUntil = 0 // Index in textSent where the audio received ends\n\n constructor(\n private readonly options: ElevenLabsTTSOptions,\n private readonly host: ElevenLabsWebSocketHost\n ) {\n // Build query params\n const params = new URLSearchParams()\n params.append('model_id', this.options.modelId ?? DEFAULT_MODEL_ID)\n params.append('output_format', getOutputFormat(this.options))\n params.append('inactivity_timeout', WS_INACTIVITY_TIMEOUT.toString())\n const voiceSettings = toWireVoiceSettings(this.options.voiceSettings)\n if (voiceSettings) {\n params.append('voice_settings', JSON.stringify(voiceSettings))\n }\n if (this.options.language) {\n params.append('language_code', this.options.language)\n }\n\n this.connection = new ElevenLabsConnection({\n url: `wss://api.elevenlabs.io/v1/text-to-speech/${this.options.voiceId}/stream-input?${params.toString()}`,\n tts: options,\n // Initialization message with voice settings\n init: { text: ' ', voice_settings: voiceSettings },\n keepAlive: {\n interval: (WS_INACTIVITY_TIMEOUT - 1) * 1000,\n message: { text: ' ' },\n },\n host,\n onOpen: () => this.resume(),\n onMessage: (message) => this.onMessage(message),\n onFailed: () => this.onFailed(),\n })\n this.connection.connect()\n }\n\n speak(textStream: Readable) {\n const generation = ++this.generation\n let started = false\n\n // Taking over is deferred until there is something to say.\n //\n // These fields track the utterance being spoken right now, and the\n // comparison between what was sent and what came back as audio is what\n // says where it is. Resetting them for a stream that never carries a word,\n // which is what an answer skipped by a tool or by onBeforeAnswer hands\n // over, loses the state of a sentence still being spoken and flushes it\n // for nothing.\n const start = () => {\n if (started) return this.speaking === generation\n // Cancelled, or superseded by another speak(), before the first word.\n if (this.generation !== generation) return false\n started = true\n // The utterance still being synthesized is superseded: drop it with\n // its connection. Otherwise connect again if the last one gave up.\n if (this.state !== 'idle') this.connection.restart()\n else this.connection.connect()\n this.resetUtterance()\n this.speaking = generation\n this.state = 'speaking'\n return true\n }\n\n // Forward text chunks coming from the caller to ElevenLabs\n textStream.on('data', async (chunk: Buffer) => {\n if (!start()) return\n this.textBuffer += chunk.toString('utf-8').replace(/[\\r\\n ]+/g, ' ')\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n\n // Only send complete words\n const spaceIndex = this.textBuffer.lastIndexOf(' ')\n if (spaceIndex === -1) return\n this.sendTranscript(this.textBuffer.slice(0, spaceIndex + 1))\n this.textBuffer = this.textBuffer.slice(spaceIndex + 1)\n })\n\n textStream.on('error', (error) => {\n if (!started || this.speaking !== generation) return\n this.log('Error in text stream, ending audio stream', error)\n // The text sent so far is still spoken, the rest is dropped\n this.speaking = 0\n this.textBuffer = ''\n this.flush()\n })\n\n textStream.on('end', async () => {\n // Nothing was ever said, so there is nothing to flush.\n if (!started) return\n if (!(await this.connection.ready())) return\n if (this.speaking !== generation) return\n // Send last buffered text\n if (this.textBuffer.trim()) {\n this.sendTranscript(this.textBuffer + ' ')\n }\n this.textBuffer = ''\n this.flush()\n })\n }\n\n cancel() {\n this.generation++\n this.speaking = 0\n if (this.state === 'idle') return\n this.log('Cancel')\n this.resetUtterance()\n this.connection.restart()\n }\n\n destroy() {\n this.generation++\n this.speaking = 0\n this.resetUtterance()\n this.connection.destroy()\n }\n\n /** Flushes buffered text and marks the end of the utterance */\n private flush() {\n this.state = 'flushed'\n this.connection.send({ text: ' ', flush: true })\n this.log('Flushed text')\n this.checkEnded()\n }\n\n /**\n * Sends the text of the utterance in flight again on a new connection. The\n * server lost what it was generating, so it goes on from the audio received.\n */\n private resume() {\n if (this.state === 'idle') return\n // Audio can stop inside a word, which goes again whole rather than from\n // its middle\n const wordStart = this.textSent.lastIndexOf(' ', this.heardUntil - 1) + 1\n const unheard = this.textSent.slice(wordStart)\n this.textSent = ''\n this.heardUntil = 0\n if (unheard.trim()) {\n this.log('Sending text chunks again')\n this.sendTranscript(unheard)\n }\n if (this.state === 'flushed') this.flush()\n }\n\n /** Every retry failed: hands the unspoken text over and drops the utterance */\n private onFailed() {\n const unspoken =\n this.state === 'idle'\n ? ''\n : this.textSent.slice(this.heardUntil) + this.textBuffer\n this.speaking = 0\n this.resetUtterance()\n this.host.failed(unspoken.trim() ? [unspoken] : [])\n }\n\n private resetUtterance() {\n this.state = 'idle'\n this.textSent = ''\n this.textBuffer = ''\n this.heardUntil = 0\n }\n\n private onMessage(message: ElevenLabsWebSocketMessage) {\n if ('audio' in message && message.audio) {\n this.processAudioMessage(message)\n }\n if ('isFinal' in message && message.isFinal && this.state !== 'idle') {\n this.log('Audio ended')\n this.resetUtterance()\n }\n // Sent before the server closes the connection, for a wrong key or an\n // unknown voice\n if ('error' in message) {\n console.error('[ElevenLabsTTS] Error:', message.error, message.message)\n }\n }\n\n private processAudioMessage(message: ElevenLabsWebSocketAudioOutputMessage) {\n const chunk = Buffer.from(message.audio, 'base64')\n this.connection.prove()\n if (this.state === 'idle') {\n this.log('Ignore audio chunk, nothing is being spoken')\n return\n }\n\n // Check text of received audio chunk\n if (message.alignment) {\n const chunkText = message.alignment.chars.join('')\n\n // A connection carries one utterance at a time, since a cancel or a\n // new utterance opens another, so its audio is always wanted. The\n // alignment only tells how far the voice got, for a reconnection. It\n // follows the text sent, a few characters aside when the voice skips\n // or rewrites some, and is placed where it fits next.\n const end = findHeardEnd(this.textSent, this.heardUntil, chunkText)\n if (end === -1) {\n this.log(`Audio chunk off the text sent (\"${chunkText}\")`)\n } else {\n this.log(`Received audio chunk (\"${chunkText}\")`)\n this.heardUntil = end\n }\n }\n\n // Send to audio stream\n this.host.audio(chunk)\n\n // ElevenLabs doesn't send isFinal after a flush, so the end of the\n // utterance is when its audio covers all its text\n this.checkEnded()\n }\n\n private checkEnded() {\n if (this.state !== 'flushed') return\n if (this.textSent.slice(this.heardUntil).trim()) return\n this.log('Audio ended')\n this.resetUtterance()\n }\n\n private sendTranscript(text: string) {\n // Counted as sent even when the connection just dropped, so the\n // reconnection sends it again\n this.textSent += text\n if (!this.connection.send({ text, try_trigger_generation: true })) return\n this.log(`Sent transcript: \"${text}\"`)\n }\n\n private log(...message: any[]) {\n this.host.log(...message)\n }\n}\n\n/**\n * Where `heard` ends in `text`, at the first place from `from` where the text\n * goes on with it, or -1 when it does nowhere\n */\nfunction findHeardEnd(text: string, from: number, heard: string): number {\n for (let start = from; start < text.length; start++) {\n const end = endOfPrefix(text, start, heard)\n if (end !== -1) return end\n }\n return -1\n}\n\n/**\n * Where `heard` ends in `text`, read from `from` and compared without\n * whitespace, or -1 when the text does not go on with it there.\n */\nfunction endOfPrefix(text: string, from: number, heard: string): number {\n let index = from\n for (let i = 0; i < heard.length; i++) {\n if (isSpace(heard[i])) continue\n while (index < text.length && isSpace(text[index])) index++\n if (text[index] !== heard[i]) return -1\n index++\n }\n return index\n}\n\nfunction isSpace(char: string) {\n return char === ' ' || char === '\\t' || char === '\\n' || char === '\\r'\n}\n\n/**\n * Voice settings as the WebSocket reads them.\n *\n * The SDK types them in camel case and the WebSocket reads snake case, so\n * similarityBoost and useSpeakerBoost would be ignored if sent as they are.\n */\nfunction toWireVoiceSettings(settings?: VoiceSettings) {\n if (!settings) return undefined\n const { similarityBoost, useSpeakerBoost, ...rest } = settings\n return {\n similarity_boost: similarityBoost,\n use_speaker_boost: useSpeakerBoost,\n ...rest,\n }\n}\n"],"mappings":";AAAA;AAAA,EACE;AAAA,EACA;AAAA,EAGA;AAAA,EACA;AAAA,EACA;AAAA,OACK;;;ACRP,SAAS,0BAA0B;AAInC,IAAM,oBAAoB;AAuCnB,IAAM,uBAAN,MAA2B;AAAA,EAOhC,YAA6B,SAAsC;AAAtC;AAH7B;AAAA;AAAA,SAAQ,SAAS;AACjB,SAAQ,WAAW;AAGjB,SAAK,SAAS,KAAK,aAAa;AAAA,EAClC;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,QAA0B;AACxB,QAAI,KAAK,OAAO,WAAW,OAAQ,QAAO,QAAQ,QAAQ,KAAK;AAC/D,WAAO,KAAK,OAAO,MAAM;AAAA,EAC3B;AAAA,EAEA,KAAK,SAA0B;AAC7B,WAAO,KAAK,OAAO,KAAK,KAAK,UAAU,OAAO,CAAC;AAAA,EACjD;AAAA;AAAA,EAGA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA,EAEA,UAAU;AACR,SAAK,OAAO,QAAQ;AAAA,EACtB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,SAAS;AACd,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,eAAe;AACrB,UAAM,EAAE,KAAK,WAAW,KAAK,IAAI,KAAK;AACtC,UAAM,SAAS,IAAI,mBAAmB;AAAA,MACpC,KAAK,KAAK,QAAQ;AAAA,MAClB,eAAe,EAAE,SAAS,EAAE,cAAc,IAAI,OAAO,EAAE;AAAA,MACvD,mBAAmB,IAAI;AAAA,MACvB,YAAY,IAAI;AAAA,MAChB,UAAU,IAAI;AAAA,MACd,WAAW;AAAA,QACT,OAAO,CAAC,OAAO,GAAG,KAAK,KAAK,UAAU,KAAK,QAAQ,IAAI,CAAC;AAAA,MAC1D;AAAA,MACA,WAAW;AAAA,QACT,UAAU,UAAU;AAAA,QACpB,SAAS,MAAM;AACb,eAAK,MAAM;AACX,iBAAO,KAAK,UAAU,UAAU,OAAO;AAAA,QACzC;AAAA,MACF;AAAA,MACA,KAAK,IAAI,YAAY,KAAK,IAAI,GAAG,OAAO;AAAA,IAC1C,CAAC;AAED,WAAO,GAAG,QAAQ,CAAC,EAAE,YAAY,MAAM;AACrC,WAAK,SAAS;AACd,WAAK,QAAQ,OAAO,WAAW;AAAA,IACjC,CAAC;AAED,WAAO,GAAG,WAAW,CAAC,SAAS;AAC7B,UAAI;AACF,aAAK,QAAQ,UAAU,KAAK,MAAM,KAAK,SAAS,CAAC,CAAC;AAAA,MACpD,SAAS,OAAY;AACnB,aAAK,IAAI,aAAa,QAAQ,MAAM,UAAU,KAAK;AACnD,aAAK,IAAI,4BAA4B,KAAK,SAAS,CAAC;AAAA,MACtD;AAAA,IACF,CAAC;AAED,WAAO,GAAG,SAAS,MAAM;AACvB,UAAI,KAAK,OAAQ;AACjB,WAAK;AACL,UAAI,KAAK,YAAY,IAAI,YAAY,mBAAoB,MAAK,OAAO;AAAA,IACvE,CAAC;AAED,WAAO,GAAG,UAAU,MAAM,KAAK,OAAO,CAAC;AACvC,WAAO;AAAA,EACT;AAAA,EAEQ,SAAS;AACf,SAAK,QAAQ,KAAK,IAAI,gCAAgC;AACtD,SAAK,WAAW;AAGhB,SAAK,OAAO,QAAQ;AACpB,SAAK,SAAS,KAAK,aAAa;AAChC,SAAK,QAAQ,SAAS;AAAA,EACxB;AACF;;;ACrGO,IAAM,mBAAmB;AACzB,IAAM,sBAA4C;AAElD,IAAM,wBAAwB;AAG9B,SAAS,gBACd,SACuC;AACvC,MAAI,QAAQ,WAAY,QAAO,OAAO,QAAQ,UAAU;AACxD,SAAO,QAAQ,gBAAgB,OAAO,mBAAmB;AAC3D;AAGO,SAAS,cAAc,SAAuC;AACnE,QAAM,QAAQ,cAAc,KAAK,gBAAgB,OAAO,CAAC;AACzD,SAAO,QAAQ,OAAO,MAAM,CAAC,CAAC,IAAI;AACpC;AAGO,SAAS,gBAAgB,SAA0B;AACxD,SAAO,QAAQ,WAAW,WAAW,KAAK,QAAQ,WAAW,WAAW;AAC1E;;;AC9CA,IAAM,sBAAsB;AAI5B,IAAM,eAAe;AAsBd,IAAM,oBAAN,MAAuD;AAAA;AAAA,EAgB5D,YACmB,SACA,MACjB;AAFiB;AACA;AAhBnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,YAAY;AACpB;AAAA,SAAQ,iBAAiB;AACzB;AAAA,SAAQ,UAAU;AAClB;AAAA,SAAQ,YAAY;AAMlB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO,OAAO,iBAAiB,gBAAgB,KAAK,OAAO,CAAC;AAC5D,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAGA,UAAM,YAAY,KAAK,QAAQ,eAAe;AAC9C,UAAM,OAAwC;AAAA,MAC5C,QAAQ,CAAC,KAAK,QAAQ,OAAO;AAAA,MAC7B,gBAAgB,cAAc,SAAY,SAAY,EAAE,UAAU;AAAA,IACpE;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,4DAA4D,OAAO,SAAS,CAAC;AAAA,MAClF,KAAK;AAAA,MACL;AAAA,MACA,WAAW;AAAA,QACT,UAAU;AAAA,QACV,SAAS,EAAE,YAAY,KAAK;AAAA,MAC9B;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAKd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,WAAK,UAAU;AACf,aAAO;AAAA,IACT;AAEA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,SAAS,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AACtD,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAClC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,SAAS,KAAK,aAAa,GAAG;AAAA,MACrC;AACA,WAAK,aAAa;AAClB,WAAK,QAAQ;AAAA,IACf,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEQ,SAAS;AACf,SAAK,YAAY;AACjB,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,OAAO,KAAK;AAClB,SAAK,WAAW;AAChB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AACtB,QAAI,KAAK,KAAK,GAAG;AACf,WAAK,IAAI,oBAAoB;AAC7B,WAAK,SAAS,IAAI;AAAA,IACpB;AACA,QAAI,KAAK,UAAU,WAAY,MAAK,MAAM;AAC1C,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SAAS,KAAK,KAAK,WAAW,KAAK;AACpD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,YAAY;AACjB,SAAK,iBAAiB;AAAA,EACxB;AAAA;AAAA,EAGQ,UAAU;AAChB,SAAK,QAAQ;AACb,SAAK,MAAM;AACX,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,UAAU,SAAoC;AACpD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,YAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,WAAK,WAAW,MAAM;AACtB,UAAI,KAAK,UAAU,QAAQ;AACzB,aAAK,IAAI,6CAA6C;AAAA,MACxD,OAAO;AACL,aAAK,IAAI,yBAAyB,MAAM,MAAM,SAAS;AACvD,aAAK,KAAK,MAAM,KAAK;AAAA,MACvB;AAAA,IACF;AACA,QACE,6BAA6B,WAC7B,QAAQ,yBACR;AACA,WAAK,iBAAiB,KAAK,IAAI,GAAG,KAAK,iBAAiB,CAAC;AACzD,WAAK,WAAW;AAAA,IAClB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA;AAAA,EAGQ,SAAS,MAAc;AAC7B,QAAI,QAAQ;AACZ,eAAW,SAAS,KAAK,SAAS,YAAY,GAAG;AAC/C,YAAM,MAAM,MAAM,QAAS,MAAM,CAAC,EAAE;AACpC,WAAK,UAAU,KAAK,MAAM,OAAO,GAAG,GAAG,IAAI;AAC3C,cAAQ;AAAA,IACV;AACA,QAAI,QAAQ,KAAK,QAAQ;AACvB,WAAK,UAAU,KAAK,MAAM,KAAK,GAAG,KAAK;AAAA,IACzC;AAAA,EACF;AAAA,EAEQ,UAAU,MAAc,OAAgB;AAC9C,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,KAAK,EAAG;AAClB,SAAK,KAAK;AAAA,MACR,QAAQ;AAAA,QACN;AAAA,UACE;AAAA,UACA,UAAU,KAAK,QAAQ;AAAA;AAAA,UAEvB,UAAW,KAAK,WAAW,KAAK,aAAc;AAAA,QAChD;AAAA,MACF;AAAA,MACA,OAAO,SAAS;AAAA,IAClB,CAAC;AACD,SAAK,IAAI,YAAY,QAAQ,iBAAiB,EAAE,MAAM,IAAI,GAAG;AAC7D,SAAK,UAAU;AACf,SAAK,YAAY;AACjB,QAAI,OAAO;AACT,WAAK;AACL,WAAK,YAAY;AAAA,IACnB,OAAO;AACL,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGQ,QAAQ;AACd,QAAI,CAAC,KAAK,UAAW;AACrB,SAAK,KAAK,EAAE,OAAO,KAAK,CAAC;AACzB,SAAK;AACL,SAAK,YAAY;AACjB,SAAK,IAAI,cAAc;AAAA,EACzB;AAAA;AAAA,EAGQ,aAAa;AACnB,QAAI,KAAK,UAAU,WAAY;AAC/B,QAAI,KAAK,aAAa,KAAK,iBAAiB,EAAG;AAC/C,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA;AAAA,EAIQ,KAAK,SAA0C;AACrD,SAAK,WAAW,KAAK,OAAO;AAAA,EAC9B;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;;;ACxRA,IAAM,wBAAwB;AAmBvB,IAAM,kBAAN,MAAqD;AAAA;AAAA,EAa1D,YACmB,SACA,MACjB;AAFiB;AACA;AAbnB,SAAQ,QAAwB;AAGhC;AAAA;AAAA,SAAQ,aAAa;AAGrB;AAAA;AAAA,SAAQ,WAAW;AACnB,SAAQ,WAAW;AACnB;AAAA,SAAQ,aAAa;AACrB;AAAA,SAAQ,aAAa;AAOnB,UAAM,SAAS,IAAI,gBAAgB;AACnC,WAAO,OAAO,YAAY,KAAK,QAAQ,WAAW,gBAAgB;AAClE,WAAO,OAAO,iBAAiB,gBAAgB,KAAK,OAAO,CAAC;AAC5D,WAAO,OAAO,sBAAsB,sBAAsB,SAAS,CAAC;AACpE,UAAM,gBAAgB,oBAAoB,KAAK,QAAQ,aAAa;AACpE,QAAI,eAAe;AACjB,aAAO,OAAO,kBAAkB,KAAK,UAAU,aAAa,CAAC;AAAA,IAC/D;AACA,QAAI,KAAK,QAAQ,UAAU;AACzB,aAAO,OAAO,iBAAiB,KAAK,QAAQ,QAAQ;AAAA,IACtD;AAEA,SAAK,aAAa,IAAI,qBAAqB;AAAA,MACzC,KAAK,6CAA6C,KAAK,QAAQ,OAAO,iBAAiB,OAAO,SAAS,CAAC;AAAA,MACxG,KAAK;AAAA;AAAA,MAEL,MAAM,EAAE,MAAM,KAAK,gBAAgB,cAAc;AAAA,MACjD,WAAW;AAAA,QACT,WAAW,wBAAwB,KAAK;AAAA,QACxC,SAAS,EAAE,MAAM,IAAI;AAAA,MACvB;AAAA,MACA;AAAA,MACA,QAAQ,MAAM,KAAK,OAAO;AAAA,MAC1B,WAAW,CAAC,YAAY,KAAK,UAAU,OAAO;AAAA,MAC9C,UAAU,MAAM,KAAK,SAAS;AAAA,IAChC,CAAC;AACD,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,MAAM,YAAsB;AAC1B,UAAM,aAAa,EAAE,KAAK;AAC1B,QAAI,UAAU;AAUd,UAAM,QAAQ,MAAM;AAClB,UAAI,QAAS,QAAO,KAAK,aAAa;AAEtC,UAAI,KAAK,eAAe,WAAY,QAAO;AAC3C,gBAAU;AAGV,UAAI,KAAK,UAAU,OAAQ,MAAK,WAAW,QAAQ;AAAA,UAC9C,MAAK,WAAW,QAAQ;AAC7B,WAAK,eAAe;AACpB,WAAK,WAAW;AAChB,WAAK,QAAQ;AACb,aAAO;AAAA,IACT;AAGA,eAAW,GAAG,QAAQ,OAAO,UAAkB;AAC7C,UAAI,CAAC,MAAM,EAAG;AACd,WAAK,cAAc,MAAM,SAAS,OAAO,EAAE,QAAQ,aAAa,GAAG;AACnE,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAGlC,YAAM,aAAa,KAAK,WAAW,YAAY,GAAG;AAClD,UAAI,eAAe,GAAI;AACvB,WAAK,eAAe,KAAK,WAAW,MAAM,GAAG,aAAa,CAAC,CAAC;AAC5D,WAAK,aAAa,KAAK,WAAW,MAAM,aAAa,CAAC;AAAA,IACxD,CAAC;AAED,eAAW,GAAG,SAAS,CAAC,UAAU;AAChC,UAAI,CAAC,WAAW,KAAK,aAAa,WAAY;AAC9C,WAAK,IAAI,6CAA6C,KAAK;AAE3D,WAAK,WAAW;AAChB,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAED,eAAW,GAAG,OAAO,YAAY;AAE/B,UAAI,CAAC,QAAS;AACd,UAAI,CAAE,MAAM,KAAK,WAAW,MAAM,EAAI;AACtC,UAAI,KAAK,aAAa,WAAY;AAElC,UAAI,KAAK,WAAW,KAAK,GAAG;AAC1B,aAAK,eAAe,KAAK,aAAa,GAAG;AAAA,MAC3C;AACA,WAAK,aAAa;AAClB,WAAK,MAAM;AAAA,IACb,CAAC;AAAA,EACH;AAAA,EAEA,SAAS;AACP,SAAK;AACL,SAAK,WAAW;AAChB,QAAI,KAAK,UAAU,OAAQ;AAC3B,SAAK,IAAI,QAAQ;AACjB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA,EAEA,UAAU;AACR,SAAK;AACL,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,WAAW,QAAQ;AAAA,EAC1B;AAAA;AAAA,EAGQ,QAAQ;AACd,SAAK,QAAQ;AACb,SAAK,WAAW,KAAK,EAAE,MAAM,KAAK,OAAO,KAAK,CAAC;AAC/C,SAAK,IAAI,cAAc;AACvB,SAAK,WAAW;AAAA,EAClB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMQ,SAAS;AACf,QAAI,KAAK,UAAU,OAAQ;AAG3B,UAAM,YAAY,KAAK,SAAS,YAAY,KAAK,KAAK,aAAa,CAAC,IAAI;AACxE,UAAM,UAAU,KAAK,SAAS,MAAM,SAAS;AAC7C,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,QAAI,QAAQ,KAAK,GAAG;AAClB,WAAK,IAAI,2BAA2B;AACpC,WAAK,eAAe,OAAO;AAAA,IAC7B;AACA,QAAI,KAAK,UAAU,UAAW,MAAK,MAAM;AAAA,EAC3C;AAAA;AAAA,EAGQ,WAAW;AACjB,UAAM,WACJ,KAAK,UAAU,SACX,KACA,KAAK,SAAS,MAAM,KAAK,UAAU,IAAI,KAAK;AAClD,SAAK,WAAW;AAChB,SAAK,eAAe;AACpB,SAAK,KAAK,OAAO,SAAS,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;AAAA,EACpD;AAAA,EAEQ,iBAAiB;AACvB,SAAK,QAAQ;AACb,SAAK,WAAW;AAChB,SAAK,aAAa;AAClB,SAAK,aAAa;AAAA,EACpB;AAAA,EAEQ,UAAU,SAAqC;AACrD,QAAI,WAAW,WAAW,QAAQ,OAAO;AACvC,WAAK,oBAAoB,OAAO;AAAA,IAClC;AACA,QAAI,aAAa,WAAW,QAAQ,WAAW,KAAK,UAAU,QAAQ;AACpE,WAAK,IAAI,aAAa;AACtB,WAAK,eAAe;AAAA,IACtB;AAGA,QAAI,WAAW,SAAS;AACtB,cAAQ,MAAM,0BAA0B,QAAQ,OAAO,QAAQ,OAAO;AAAA,IACxE;AAAA,EACF;AAAA,EAEQ,oBAAoB,SAAgD;AAC1E,UAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,SAAK,WAAW,MAAM;AACtB,QAAI,KAAK,UAAU,QAAQ;AACzB,WAAK,IAAI,6CAA6C;AACtD;AAAA,IACF;AAGA,QAAI,QAAQ,WAAW;AACrB,YAAM,YAAY,QAAQ,UAAU,MAAM,KAAK,EAAE;AAOjD,YAAM,MAAM,aAAa,KAAK,UAAU,KAAK,YAAY,SAAS;AAClE,UAAI,QAAQ,IAAI;AACd,aAAK,IAAI,mCAAmC,SAAS,IAAI;AAAA,MAC3D,OAAO;AACL,aAAK,IAAI,0BAA0B,SAAS,IAAI;AAChD,aAAK,aAAa;AAAA,MACpB;AAAA,IACF;AAGA,SAAK,KAAK,MAAM,KAAK;AAIrB,SAAK,WAAW;AAAA,EAClB;AAAA,EAEQ,aAAa;AACnB,QAAI,KAAK,UAAU,UAAW;AAC9B,QAAI,KAAK,SAAS,MAAM,KAAK,UAAU,EAAE,KAAK,EAAG;AACjD,SAAK,IAAI,aAAa;AACtB,SAAK,eAAe;AAAA,EACtB;AAAA,EAEQ,eAAe,MAAc;AAGnC,SAAK,YAAY;AACjB,QAAI,CAAC,KAAK,WAAW,KAAK,EAAE,MAAM,wBAAwB,KAAK,CAAC,EAAG;AACnE,SAAK,IAAI,qBAAqB,IAAI,GAAG;AAAA,EACvC;AAAA,EAEQ,OAAO,SAAgB;AAC7B,SAAK,KAAK,IAAI,GAAG,OAAO;AAAA,EAC1B;AACF;AAMA,SAAS,aAAa,MAAc,MAAc,OAAuB;AACvE,WAAS,QAAQ,MAAM,QAAQ,KAAK,QAAQ,SAAS;AACnD,UAAM,MAAM,YAAY,MAAM,OAAO,KAAK;AAC1C,QAAI,QAAQ,GAAI,QAAO;AAAA,EACzB;AACA,SAAO;AACT;AAMA,SAAS,YAAY,MAAc,MAAc,OAAuB;AACtE,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,QAAI,QAAQ,MAAM,CAAC,CAAC,EAAG;AACvB,WAAO,QAAQ,KAAK,UAAU,QAAQ,KAAK,KAAK,CAAC,EAAG;AACpD,QAAI,KAAK,KAAK,MAAM,MAAM,CAAC,EAAG,QAAO;AACrC;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,QAAQ,MAAc;AAC7B,SAAO,SAAS,OAAO,SAAS,OAAQ,SAAS,QAAQ,SAAS;AACpE;AAQA,SAAS,oBAAoB,UAA0B;AACrD,MAAI,CAAC,SAAU,QAAO;AACtB,QAAM,EAAE,iBAAiB,iBAAiB,GAAG,KAAK,IAAI;AACtD,SAAO;AAAA,IACL,kBAAkB;AAAA,IAClB,mBAAmB;AAAA,IACnB,GAAG;AAAA,EACL;AACF;;;AJlSA,IAAM,oBAAoB,CAAC,UAAU,aAAa,cAAc;AASzD,IAAM,gBAAN,cAA4B,IAAI;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOrC,OAAO,WAAW,SAGU;AAC1B,WAAO;AAAA,MACL,CAAC,YAAY;AAAA,MACb,QAAQ;AAAA,MACR,YAAY;AACV,cAAM,WAAW,MAAM,MAAM,uCAAuC;AAAA,UAClE,SAAS,EAAE,cAAc,QAAQ,OAAO;AAAA,QAC1C,CAAC;AACD,YAAI,CAAC,SAAS,IAAI;AAChB,gBAAM,IAAI,MAAM,uBAAuB,SAAS,MAAM,EAAE;AAAA,QAC1D;AACA,cAAM,SAAU,MAAM,SAAS,KAAK;AAOpC,eAAO,OACJ,OAAO,CAAC,UAAU,MAAM,qBAAqB,EAC7C,IAAI,CAAC,WAAW;AAAA,UACf,IAAI,MAAM;AAAA,UACV,MAAM,MAAM;AAAA,UACZ,aAAa,MAAM;AAAA,UACnB,WAAW,MAAM,WAAW,IAAI,CAAC,EAAE,YAAY,MAAM,WAAW;AAAA,QAClE,EAAE;AAAA,MACN;AAAA,MACA,QAAQ;AAAA,IACV;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAWA,OAAO,WAAW,SAIU;AAC1B,WAAO;AAAA,MACL,CAAC,YAAY;AAAA,MACb,QAAQ;AAAA,MACR,MAAM,YAAY,QAAQ,MAAM;AAAA,MAChC,QAAQ;AAAA,IACV,EAAE,KAAK,CAAC,WAAW,kBAAkB,QAAQ,QAAQ,QAAQ,CAAC;AAAA,EAChE;AAAA,EAKA,YAAY,SAA+B;AACzC,UAAM;AACN,SAAK,OAAO,cAAc,OAAO;AACjC,UAAM,OAAO;AAAA,MACX,OAAO,CAAC,UAAkB,KAAK,KAAK,SAAS,KAAK;AAAA,MAClD,QAAQ,CAAC,UAAoB,KAAK,KAAK,UAAU,KAAK;AAAA,MACtD,KAAK,IAAI,YAAmB,KAAK,IAAI,GAAG,OAAO;AAAA,IACjD;AACA,SAAK,YAAY,gBAAgB,QAAQ,WAAW,gBAAgB,IAChE,IAAI,kBAAkB,SAAS,IAAI,IACnC,IAAI,gBAAgB,SAAS,IAAI;AAAA,EACvC;AAAA,EAEA,IAAI,aAAa;AACf,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAM,YAAsB;AAC1B,SAAK,UAAU,MAAM,UAAU;AAAA,EACjC;AAAA,EAEA,SAAS;AACP,SAAK,UAAU,OAAO;AAAA,EACxB;AAAA,EAEA,UAAU;AACR,SAAK,UAAU,QAAQ;AACvB,UAAM,QAAQ;AAAA,EAChB;AACF;AAGA,eAAe,YAAY,QAAyC;AAClE,QAAM,SAAyB,CAAC;AAChC,MAAI;AACJ,KAAG;AACD,UAAM,SAAS,IAAI,gBAAgB,EAAE,WAAW,MAAM,CAAC;AACvD,QAAI,UAAW,QAAO,IAAI,mBAAmB,SAAS;AACtD,UAAM,WAAW,MAAM;AAAA,MACrB,uCAAuC,MAAM;AAAA,MAC7C,EAAE,SAAS,EAAE,cAAc,OAAO,EAAE;AAAA,IACtC;AACA,QAAI,CAAC,SAAS,GAAI,OAAM,IAAI,MAAM,uBAAuB,SAAS,MAAM,EAAE;AAC1E,UAAM,OAAQ,MAAM,SAAS,KAAK;AAKlC,WAAO,KAAK,GAAG,KAAK,OAAO,IAAI,cAAc,CAAC;AAC9C,gBAAY,KAAK,WAAY,KAAK,mBAAmB,SAAa;AAAA,EACpE,SAAS;AACT,SAAO;AACT;AAEA,SAAS,eAAe,OAAsC;AAC5D,QAAM,YAAY,MAAM,sBAAsB,CAAC,GAC5C,IAAI,CAAC,EAAE,QAAQ,SAAS,MAAM,UAAU,QAAQ,EAChD,OAAO,CAAC,SAAyB,CAAC,CAAC,IAAI;AAC1C,QAAM,QAAQ,MAAM,QAAQ;AAC5B,QAAM,YAAY,SAAS,SAAS,WAAW,QAAQ,CAAC,KAAK,IAAI;AACjE,SAAO;AAAA,IACL,IAAI,MAAM;AAAA,IACV,MAAM,MAAM;AAAA,IACZ,WAAW,aAAa,CAAC,GAAG,IAAI,IAAI,SAAS,CAAC;AAAA,IAC9C,QAAQ,YAAY,MAAM,QAAQ,MAAM;AAAA,IACxC,aAAa,MAAM,eAAe;AAAA,IAClC,YAAY,MAAM,eAAe;AAAA,IACjC,QAAQ,kBAAkB,SAAS,MAAM,YAAY,EAAE;AAAA,EACzD;AACF;","names":[]}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@micdrop/elevenlabs",
|
|
3
|
-
"version": "1.
|
|
3
|
+
"version": "1.4.0",
|
|
4
4
|
"description": "ElevenLabs implementation for @micdrop/server",
|
|
5
5
|
"author": "Godefroy de Compreignac",
|
|
6
6
|
"license": "MIT",
|
|
@@ -39,7 +39,7 @@
|
|
|
39
39
|
"ws": "^8.19.0"
|
|
40
40
|
},
|
|
41
41
|
"peerDependencies": {
|
|
42
|
-
"@micdrop/server": ">=2.
|
|
42
|
+
"@micdrop/server": ">=2.11.0"
|
|
43
43
|
},
|
|
44
44
|
"devDependencies": {
|
|
45
45
|
"@types/node": "^22.19.13",
|