@jskit-ai/assistant-voice-sherpa 0.1.41 → 0.1.43
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/models/voice-models-kitten.json +1 -1
- package/models/voice-models-kokoro-q8f16.json +1 -1
- package/models/voice-models-kokoro.json +1 -1
- package/models/voice-models-piper.json +1 -1
- package/models/voice-models.json +1 -1
- package/package.json +2 -2
- package/src/cli.js +36 -20
- package/src/sherpaSpeechEngine.js +2 -1
- package/src/voiceModelContract.js +1 -1
- package/tooling/prepare-models.sh +4 -2
package/models/voice-models.json
CHANGED
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@jskit-ai/assistant-voice-sherpa",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.43",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"description": "Optional native Sherpa speech engine and explicit model preparation for assistant voice.",
|
|
6
6
|
"scripts": {
|
|
@@ -19,7 +19,7 @@
|
|
|
19
19
|
"tooling"
|
|
20
20
|
],
|
|
21
21
|
"dependencies": {
|
|
22
|
-
"@jskit-ai/assistant-voice": "0.1.
|
|
22
|
+
"@jskit-ai/assistant-voice": "0.1.42",
|
|
23
23
|
"sherpa-onnx-node": "1.13.6"
|
|
24
24
|
},
|
|
25
25
|
"jskit": {
|
package/src/cli.js
CHANGED
|
@@ -9,6 +9,33 @@ import { createVoiceAccessToken } from "@jskit-ai/assistant-voice/server";
|
|
|
9
9
|
import { createVoiceDaemon } from "@jskit-ai/assistant-voice/server";
|
|
10
10
|
import { verifyVoiceModelContract } from "./voiceModelContract.js";
|
|
11
11
|
|
|
12
|
+
// Combined packs list their manifests in merge order: base, kitten, kokoro.
|
|
13
|
+
const PACKS = Object.freeze({
|
|
14
|
+
cori: ["voice-models.json"],
|
|
15
|
+
piper: ["voice-models-piper.json"],
|
|
16
|
+
kitten: ["voice-models-kitten.json"],
|
|
17
|
+
"piper-kitten": ["voice-models-piper.json", "voice-models-kitten.json"],
|
|
18
|
+
kokoro: ["voice-models-kokoro.json"],
|
|
19
|
+
"kokoro-q8f16": ["voice-models-kokoro-q8f16.json"],
|
|
20
|
+
"piper-kitten-kokoro-q8f16": ["voice-models-piper.json", "voice-models-kitten.json", "voice-models-kokoro-q8f16.json"]
|
|
21
|
+
});
|
|
22
|
+
const PACK_NAMES = Object.keys(PACKS);
|
|
23
|
+
|
|
24
|
+
function mergeVoiceModelSources([base, kitten, kokoro]) {
|
|
25
|
+
base.models.push(...kitten.models.filter(model => model.kind === "tts"));
|
|
26
|
+
base.configuration.voices.unshift(...kitten.configuration.voices);
|
|
27
|
+
Object.assign(base.configuration.synthesizers, kitten.configuration.synthesizers);
|
|
28
|
+
if (kokoro) {
|
|
29
|
+
base.models.push(...kokoro.models.filter(model => model.kind === "tts"));
|
|
30
|
+
base.configuration.voices.push(...kokoro.configuration.voices.map(voice => ({ ...voice, modelId: "kokoro" })));
|
|
31
|
+
base.configuration.synthesizers.kokoro = kokoro.configuration.synthesizer;
|
|
32
|
+
base.onnxMetadata = kokoro.onnxMetadata;
|
|
33
|
+
base.textFiles = kokoro.textFiles;
|
|
34
|
+
}
|
|
35
|
+
base.configuration.outputSampleRate = 22050;
|
|
36
|
+
return base;
|
|
37
|
+
}
|
|
38
|
+
|
|
12
39
|
function argumentMap(values = []) {
|
|
13
40
|
const [command = "help", ...tail] = values;
|
|
14
41
|
const options = {};
|
|
@@ -50,7 +77,7 @@ function usage() {
|
|
|
50
77
|
"",
|
|
51
78
|
"Commands:",
|
|
52
79
|
" serve --models-root PATH --key-file PATH [--host ADDRESS] [--port NUMBER] [--config FILE]",
|
|
53
|
-
|
|
80
|
+
` prepare --models-root PATH [--pack ${PACK_NAMES.join("|")}] [--sources-file FILE] [--download-cache PATH] [--hotwords-file PATH]`,
|
|
54
81
|
" verify --models-root PATH",
|
|
55
82
|
" token --key-file PATH --tenant NAME"
|
|
56
83
|
].join("\n");
|
|
@@ -77,31 +104,20 @@ async function runVoiceCli(argv = process.argv.slice(2), { createDaemon = create
|
|
|
77
104
|
"retain-downloads": "RETAIN_DOWNLOADS", "hotwords-file": "HOTWORDS_FILE"
|
|
78
105
|
})) if (options[option]) env[`JSKIT_VOICE_${variable}`] = options[option];
|
|
79
106
|
const pack = options.pack || process.env.JSKIT_VOICE_PACK || "cori";
|
|
80
|
-
|
|
81
|
-
|
|
107
|
+
if (!Object.hasOwn(PACKS, pack)) throw new Error(`Choose --pack ${PACK_NAMES.slice(0, -1).join(", ")} or ${PACK_NAMES.at(-1)}, or supply --sources-file for your own model pack.`);
|
|
108
|
+
const manifests = PACKS[pack];
|
|
82
109
|
let temporary;
|
|
83
110
|
try {
|
|
84
111
|
env.JSKIT_VOICE_SOURCES_FILE = options["sources-file"] || process.env.JSKIT_VOICE_SOURCES_FILE;
|
|
85
|
-
if (!env.JSKIT_VOICE_SOURCES_FILE &&
|
|
86
|
-
const
|
|
87
|
-
|
|
88
|
-
|
|
89
|
-
piper.configuration.voices.unshift(...kitten.configuration.voices);
|
|
90
|
-
Object.assign(piper.configuration.synthesizers, kitten.configuration.synthesizers);
|
|
91
|
-
if (pack === "piper-kitten-kokoro-q8f16") {
|
|
92
|
-
const kokoro = JSON.parse(await readFile(new URL("../models/voice-models-kokoro-q8f16.json", import.meta.url), "utf8"));
|
|
93
|
-
piper.models.push(...kokoro.models.filter(model => model.kind === "tts"));
|
|
94
|
-
piper.configuration.voices.push(...kokoro.configuration.voices.map(voice => ({ ...voice, modelId: "kokoro" })));
|
|
95
|
-
piper.configuration.synthesizers.kokoro = kokoro.configuration.synthesizer;
|
|
96
|
-
piper.onnxMetadata = kokoro.onnxMetadata;
|
|
97
|
-
piper.textFiles = kokoro.textFiles;
|
|
98
|
-
}
|
|
99
|
-
piper.configuration.outputSampleRate = 22050;
|
|
112
|
+
if (!env.JSKIT_VOICE_SOURCES_FILE && manifests.length > 1) {
|
|
113
|
+
const sources = mergeVoiceModelSources(await Promise.all(manifests.map(async manifest => (
|
|
114
|
+
JSON.parse(await readFile(new URL(`../models/${manifest}`, import.meta.url), "utf8"))
|
|
115
|
+
))));
|
|
100
116
|
temporary = await mkdtemp(path.join(os.tmpdir(), "jskit-voice-sources-"));
|
|
101
117
|
env.JSKIT_VOICE_SOURCES_FILE = path.join(temporary, "sources.json");
|
|
102
|
-
await writeFile(env.JSKIT_VOICE_SOURCES_FILE, JSON.stringify(
|
|
118
|
+
await writeFile(env.JSKIT_VOICE_SOURCES_FILE, JSON.stringify(sources, null, 2) + "\n");
|
|
103
119
|
}
|
|
104
|
-
env.JSKIT_VOICE_SOURCES_FILE ||= fileURLToPath(new URL(`../models/${manifests[
|
|
120
|
+
env.JSKIT_VOICE_SOURCES_FILE ||= fileURLToPath(new URL(`../models/${manifests[0]}`, import.meta.url));
|
|
105
121
|
await new Promise((resolve, reject) => {
|
|
106
122
|
const child = spawn("bash", [script], { env, stdio: "inherit" });
|
|
107
123
|
child.once("error", reject);
|
|
@@ -3,6 +3,7 @@ import { createRequire } from "node:module";
|
|
|
3
3
|
import { createSynthesisProcess, SYNTHESIS_REQUEST_TIMEOUT_MS } from "./synthesisProcess.js";
|
|
4
4
|
|
|
5
5
|
import {
|
|
6
|
+
VOICE_ID_PATTERN,
|
|
6
7
|
VOICE_INPUT_SAMPLE_RATE,
|
|
7
8
|
normalizeRecognizedText,
|
|
8
9
|
pcm16LeToFloat32
|
|
@@ -105,7 +106,7 @@ async function createSherpaSpeechEngine({
|
|
|
105
106
|
? [{ id: "default", label: "Default voice", speakerId: 0 }]
|
|
106
107
|
: [{ id: "cori", label: "Cori · female · British English · Piper", language: "en-GB", speakerId: 0 }]);
|
|
107
108
|
if (!Array.isArray(voices) || !voices.length || voices.length > 128 || new Set(voices.map(voice => voice?.id)).size !== voices.length ||
|
|
108
|
-
voices.some(voice => !voice ||
|
|
109
|
+
voices.some(voice => !voice || !VOICE_ID_PATTERN.test(voice.id) || typeof voice.label !== "string" || !voice.label.trim() || voice.label.length > 100 ||
|
|
109
110
|
!Number.isSafeInteger(voice.speakerId) || voice.speakerId < 0)) {
|
|
110
111
|
throw new TypeError("Configure unique voice IDs, names and valid model speaker IDs.");
|
|
111
112
|
}
|
|
@@ -3,7 +3,7 @@ import { createReadStream } from "node:fs";
|
|
|
3
3
|
import { readFile, stat } from "node:fs/promises";
|
|
4
4
|
import path from "node:path";
|
|
5
5
|
|
|
6
|
-
const VOICE_MODEL_CONTRACT_SCHEMA = "
|
|
6
|
+
const VOICE_MODEL_CONTRACT_SCHEMA = "jskit.assistant-voice.models.v1";
|
|
7
7
|
const SHA256_PATTERN = /^[a-f0-9]{64}$/u;
|
|
8
8
|
|
|
9
9
|
function modelContractError(code, message) {
|
|
@@ -73,7 +73,9 @@ if [ -f "$models_root/voice-models.json" ] &&
|
|
|
73
73
|
cmp -s "$hotwords_source" "$models_root/stt/hotwords.txt" &&
|
|
74
74
|
cmp -s "$bpe_vocab_source" "$models_root/stt/bpe.vocab" &&
|
|
75
75
|
"$node_bin" "$model_contract_verifier" "$models_root"; then
|
|
76
|
-
|
|
76
|
+
# The speech service runs as its own user, so the pack is readable by all
|
|
77
|
+
# whatever umask prepared it.
|
|
78
|
+
chmod -R u=rwX,go=rX "$models_root"
|
|
77
79
|
if [ "$retain_downloads" = "0" ]; then
|
|
78
80
|
while IFS= read -r model_id; do
|
|
79
81
|
rm -f -- "$download_cache/$(read_model_field "$model_id" archive)"
|
|
@@ -141,7 +143,7 @@ cp -- "$source_manifest" "$stage_root/sources.json"
|
|
|
141
143
|
|
|
142
144
|
"$node_bin" "$model_contract_builder" "$stage_root"
|
|
143
145
|
"$node_bin" "$model_contract_verifier" "$stage_root"
|
|
144
|
-
chmod
|
|
146
|
+
chmod -R u=rwX,go=rX "$stage_root"
|
|
145
147
|
|
|
146
148
|
rm -rf -- "$previous_root"
|
|
147
149
|
if [[ -e "$models_root" ]]; then
|