@fonoster/apiserver 0.26.0 → 0.27.0
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.
|
@@ -20,7 +20,7 @@ import { Stream } from "stream";
|
|
|
20
20
|
import * as z from "zod";
|
|
21
21
|
import { SpeechToText } from "../types";
|
|
22
22
|
import { AbstractSpeechToText } from "./AbstractSpeechToText";
|
|
23
|
-
import { DeepgramSttConfig, SpeechResult, StreamSpeech } from "./types";
|
|
23
|
+
import { DeepgramLanguage, DeepgramModel, DeepgramSttConfig, SpeechResult, StreamSpeech } from "./types";
|
|
24
24
|
declare const DeepgramClient: any;
|
|
25
25
|
declare const ENGINE_NAME = "stt.deepgram";
|
|
26
26
|
declare class Deepgram extends AbstractSpeechToText<typeof ENGINE_NAME> implements SpeechToText {
|
|
@@ -33,4 +33,28 @@ declare class Deepgram extends AbstractSpeechToText<typeof ENGINE_NAME> implemen
|
|
|
33
33
|
static getConfigValidationSchema(): z.Schema;
|
|
34
34
|
static getCredentialsValidationSchema(): z.Schema;
|
|
35
35
|
}
|
|
36
|
-
|
|
36
|
+
declare function buildTranscribeConfig(config: {
|
|
37
|
+
model: DeepgramModel;
|
|
38
|
+
languageCode: DeepgramLanguage;
|
|
39
|
+
smartFormat?: boolean;
|
|
40
|
+
noDelay?: boolean;
|
|
41
|
+
interimResults?: boolean;
|
|
42
|
+
utteranceEndMs?: number;
|
|
43
|
+
}): {
|
|
44
|
+
endpointing: number;
|
|
45
|
+
model: DeepgramModel;
|
|
46
|
+
language: DeepgramLanguage;
|
|
47
|
+
encoding: string;
|
|
48
|
+
sample_rate: number;
|
|
49
|
+
channels: number;
|
|
50
|
+
smart_format: boolean;
|
|
51
|
+
no_delay: boolean;
|
|
52
|
+
interim_results: boolean;
|
|
53
|
+
utterance_end_ms: number;
|
|
54
|
+
languageCode: DeepgramLanguage;
|
|
55
|
+
smartFormat?: boolean;
|
|
56
|
+
noDelay?: boolean;
|
|
57
|
+
interimResults?: boolean;
|
|
58
|
+
utteranceEndMs?: number;
|
|
59
|
+
};
|
|
60
|
+
export { buildTranscribeConfig, Deepgram, ENGINE_NAME };
|
|
@@ -43,6 +43,7 @@ var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, ge
|
|
|
43
43
|
};
|
|
44
44
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
45
45
|
exports.ENGINE_NAME = exports.Deepgram = void 0;
|
|
46
|
+
exports.buildTranscribeConfig = buildTranscribeConfig;
|
|
46
47
|
/**
|
|
47
48
|
* Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
|
|
48
49
|
* http://github.com/fonoster/fonoster
|
|
@@ -72,6 +73,11 @@ const { DeepgramClient, LiveTranscriptionEvents, createClient
|
|
|
72
73
|
} = require("@deepgram/sdk"); // Why Deepgram :(
|
|
73
74
|
const ENGINE_NAME = "stt.deepgram";
|
|
74
75
|
exports.ENGINE_NAME = ENGINE_NAME;
|
|
76
|
+
// Nova-3 detects English, Spanish and eight other languages; Nova-2 only
|
|
77
|
+
// English + Spanish. The phonecall/conversationalai variants are English-only.
|
|
78
|
+
const MULTILINGUAL_MODELS = [types_1.DeepgramModel.NOVA_3, types_1.DeepgramModel.NOVA_2];
|
|
79
|
+
// Deepgram's recommended endpointing for code-switching (its default is 10ms)
|
|
80
|
+
const MULTILINGUAL_ENDPOINTING_MS = 100;
|
|
75
81
|
const logger = (0, logger_1.getLogger)({ service: "apiserver", filePath: __filename });
|
|
76
82
|
class Deepgram extends AbstractSpeechToText_1.AbstractSpeechToText {
|
|
77
83
|
constructor(config) {
|
|
@@ -269,19 +275,24 @@ class Deepgram extends AbstractSpeechToText_1.AbstractSpeechToText {
|
|
|
269
275
|
});
|
|
270
276
|
}
|
|
271
277
|
static getConfigValidationSchema() {
|
|
272
|
-
return z
|
|
278
|
+
return z
|
|
279
|
+
.object({
|
|
273
280
|
smartFormat: z.boolean().optional(),
|
|
274
281
|
noDelay: z.boolean().optional(),
|
|
275
282
|
languageCode: z
|
|
276
|
-
.nativeEnum(common_1.VoiceLanguage, {
|
|
277
|
-
message: common_1.Messages.VALID_LANGUAGE_CODE
|
|
278
|
-
})
|
|
283
|
+
.union([z.nativeEnum(common_1.VoiceLanguage), z.literal(types_1.DEEPGRAM_MULTILINGUAL)], { errorMap: () => ({ message: common_1.Messages.VALID_LANGUAGE_CODE }) })
|
|
279
284
|
.optional(),
|
|
280
285
|
model: z
|
|
281
286
|
.nativeEnum(types_1.DeepgramModel, { message: "Invalid Deepgram model" })
|
|
282
287
|
.optional(),
|
|
283
288
|
interimResults: z.boolean().optional(),
|
|
284
289
|
utteranceEndMs: z.number().int().min(1000).max(5000).optional()
|
|
290
|
+
})
|
|
291
|
+
.refine((config) => config.languageCode !== types_1.DEEPGRAM_MULTILINGUAL ||
|
|
292
|
+
!config.model ||
|
|
293
|
+
MULTILINGUAL_MODELS.includes(config.model), {
|
|
294
|
+
message: `Language "${types_1.DEEPGRAM_MULTILINGUAL}" requires one of these models: ${MULTILINGUAL_MODELS.join(", ")}`,
|
|
295
|
+
path: ["model"]
|
|
285
296
|
});
|
|
286
297
|
}
|
|
287
298
|
static getCredentialsValidationSchema() {
|
|
@@ -298,11 +309,15 @@ function buildTranscribeConfig(config) {
|
|
|
298
309
|
// Default utterance_end_ms to 1000ms (minimum required value)
|
|
299
310
|
// This enables UtteranceEnd events as a fallback when speech_final never becomes true
|
|
300
311
|
const utteranceEndMs = config.utteranceEndMs || 1000;
|
|
301
|
-
|
|
312
|
+
const isMultilingual = config.languageCode === types_1.DEEPGRAM_MULTILINGUAL;
|
|
313
|
+
return Object.assign(Object.assign(Object.assign({}, config), {
|
|
314
|
+
// The phonecall default is English-only, so multilingual needs its own
|
|
315
|
+
model: config.model ||
|
|
316
|
+
(isMultilingual ? types_1.DeepgramModel.NOVA_3 : types_1.DeepgramModel.NOVA_2_PHONECALL), language: config.languageCode || common_1.VoiceLanguage.EN_US, encoding: "linear16", sample_rate: 16000, channels: 1, smart_format: config.smartFormat !== false,
|
|
302
317
|
// This needs to be set to true to avoid delays while using smart_format
|
|
303
318
|
no_delay: config.noDelay !== false,
|
|
304
319
|
// REQUIRED for UtteranceEnd: interim_results must be true
|
|
305
320
|
interim_results: interimResults,
|
|
306
321
|
// REQUIRED for UtteranceEnd: utterance_end_ms parameter
|
|
307
|
-
utterance_end_ms: utteranceEndMs });
|
|
322
|
+
utterance_end_ms: utteranceEndMs }), (isMultilingual && { endpointing: MULTILINGUAL_ENDPOINTING_MS }));
|
|
308
323
|
}
|
|
@@ -27,7 +27,7 @@ type StreamSpeech = {
|
|
|
27
27
|
};
|
|
28
28
|
type SttConfig = {
|
|
29
29
|
config: {
|
|
30
|
-
languageCode:
|
|
30
|
+
languageCode: string;
|
|
31
31
|
};
|
|
32
32
|
};
|
|
33
33
|
type GoogleSttConfig = {
|
|
@@ -45,9 +45,11 @@ declare enum DeepgramModel {
|
|
|
45
45
|
NOVA_2_PHONECALL = "nova-2-phonecall",
|
|
46
46
|
NOVA_2_CONVERSATIONALAI = "nova-2-conversationalai"
|
|
47
47
|
}
|
|
48
|
+
declare const DEEPGRAM_MULTILINGUAL = "multi";
|
|
49
|
+
type DeepgramLanguage = VoiceLanguage | typeof DEEPGRAM_MULTILINGUAL;
|
|
48
50
|
type DeepgramSttConfig = {
|
|
49
51
|
config: {
|
|
50
|
-
languageCode:
|
|
52
|
+
languageCode: DeepgramLanguage;
|
|
51
53
|
model: DeepgramModel;
|
|
52
54
|
smartFormat: boolean;
|
|
53
55
|
noDelay: boolean;
|
|
@@ -58,4 +60,4 @@ type DeepgramSttConfig = {
|
|
|
58
60
|
apiKey: string;
|
|
59
61
|
};
|
|
60
62
|
};
|
|
61
|
-
export { DeepgramModel, DeepgramSttConfig, GoogleSttConfig, SpeechResult, StreamSpeech, SttConfig };
|
|
63
|
+
export { DEEPGRAM_MULTILINGUAL, DeepgramLanguage, DeepgramModel, DeepgramSttConfig, GoogleSttConfig, SpeechResult, StreamSpeech, SttConfig };
|
package/dist/voice/stt/types.js
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
"use strict";
|
|
2
2
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
-
exports.DeepgramModel = void 0;
|
|
3
|
+
exports.DeepgramModel = exports.DEEPGRAM_MULTILINGUAL = void 0;
|
|
4
4
|
var DeepgramModel;
|
|
5
5
|
(function (DeepgramModel) {
|
|
6
6
|
DeepgramModel["NOVA_3"] = "nova-3";
|
|
@@ -8,3 +8,7 @@ var DeepgramModel;
|
|
|
8
8
|
DeepgramModel["NOVA_2_PHONECALL"] = "nova-2-phonecall";
|
|
9
9
|
DeepgramModel["NOVA_2_CONVERSATIONALAI"] = "nova-2-conversationalai";
|
|
10
10
|
})(DeepgramModel || (exports.DeepgramModel = DeepgramModel = {}));
|
|
11
|
+
// Deepgram's code-switching mode: transcribes speech that mixes languages
|
|
12
|
+
// (e.g. a caller who answers "Hello" on a Spanish call). Nova-3 and Nova-2.
|
|
13
|
+
const DEEPGRAM_MULTILINGUAL = "multi";
|
|
14
|
+
exports.DEEPGRAM_MULTILINGUAL = DEEPGRAM_MULTILINGUAL;
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@fonoster/apiserver",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.27.0",
|
|
4
4
|
"description": "APIServer for Fonoster",
|
|
5
5
|
"author": "Pedro Sanders <psanders@fonoster.com>",
|
|
6
6
|
"homepage": "https://github.com/fonoster/fonoster#readme",
|
|
@@ -22,14 +22,14 @@
|
|
|
22
22
|
},
|
|
23
23
|
"dependencies": {
|
|
24
24
|
"@deepgram/sdk": "^3.5.1",
|
|
25
|
-
"@fonoster/authz": "^0.
|
|
26
|
-
"@fonoster/autopilot": "^0.
|
|
27
|
-
"@fonoster/common": "^0.
|
|
28
|
-
"@fonoster/identity": "^0.
|
|
29
|
-
"@fonoster/logger": "^0.
|
|
30
|
-
"@fonoster/sipnet": "^0.
|
|
31
|
-
"@fonoster/streams": "^0.
|
|
32
|
-
"@fonoster/types": "^0.
|
|
25
|
+
"@fonoster/authz": "^0.27.0",
|
|
26
|
+
"@fonoster/autopilot": "^0.27.0",
|
|
27
|
+
"@fonoster/common": "^0.27.0",
|
|
28
|
+
"@fonoster/identity": "^0.27.0",
|
|
29
|
+
"@fonoster/logger": "^0.27.0",
|
|
30
|
+
"@fonoster/sipnet": "^0.27.0",
|
|
31
|
+
"@fonoster/streams": "^0.27.0",
|
|
32
|
+
"@fonoster/types": "^0.27.0",
|
|
33
33
|
"@google-cloud/speech": "^6.6.0",
|
|
34
34
|
"@google-cloud/text-to-speech": "^5.3.0",
|
|
35
35
|
"@grpc/grpc-js": "~1.10.11",
|
|
@@ -76,7 +76,7 @@
|
|
|
76
76
|
"@types/uuid": "^10.0.0",
|
|
77
77
|
"@types/validator": "^13.15.10"
|
|
78
78
|
},
|
|
79
|
-
"gitHead": "
|
|
79
|
+
"gitHead": "1191b34677b51a36000177887ff82303ef2693e1",
|
|
80
80
|
"optionalDependencies": {
|
|
81
81
|
"@ai-coustics/aic-sdk": "^0.24.0"
|
|
82
82
|
}
|