dsh-voice 0.8.1 → 0.9.1
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/CHANGELOG.md +18 -0
- package/README.en.md +17 -15
- package/README.md +17 -15
- package/docs/USAGE.en.md +39 -16
- package/docs/USAGE.md +41 -18
- package/lib/client.js +5067 -0
- package/lib/config.d.ts +7 -0
- package/lib/config.js +21 -0
- package/lib/edge-tts.d.ts +4 -0
- package/lib/edge-tts.js +38 -1
- package/lib/host-config.js +7 -0
- package/lib/index.d.ts +11 -10
- package/lib/index.js +81 -1
- package/lib/openai-tts.d.ts +19 -6
- package/lib/openai-tts.js +163 -2
- package/lib/stt.d.ts +19 -0
- package/lib/stt.js +157 -9
- package/lib/tools.d.ts +4 -2
- package/lib/tools.js +380 -70
- package/lib/tts-text.d.ts +46 -3
- package/lib/tts-text.js +345 -16
- package/lib/web.d.ts +66 -0
- package/lib/web.js +691 -0
- package/locale/en.json +1 -1
- package/locale/zh.json +1 -1
- package/package.json +10 -5
- package/skills/voice-workflow/SKILL.md +51 -19
- package/lib/update-client.js +0 -745
package/lib/stt.js
CHANGED
|
@@ -68,7 +68,7 @@ export async function transcribe(baseUrl, apiKey, options, fetchImpl = globalThi
|
|
|
68
68
|
form.append('language', options.language);
|
|
69
69
|
if (options.prompt !== undefined && options.prompt !== '')
|
|
70
70
|
form.append('prompt', options.prompt);
|
|
71
|
-
if (options.timestamps === true)
|
|
71
|
+
if (options.timestamps === true || options.diarize === true)
|
|
72
72
|
form.append('response_format', 'verbose_json');
|
|
73
73
|
const headers = apiKey !== '' ? { authorization: 'Bearer ' + apiKey } : {};
|
|
74
74
|
const timeoutSignal = AbortSignal.timeout(timeoutMs);
|
|
@@ -134,19 +134,47 @@ export async function transcribe(baseUrl, apiKey, options, fetchImpl = globalThi
|
|
|
134
134
|
if (Array.isArray(rec.segments)) {
|
|
135
135
|
segments = rec.segments
|
|
136
136
|
.filter((item) => typeof item === 'object' && item !== null)
|
|
137
|
-
.map((item, idx) =>
|
|
138
|
-
|
|
139
|
-
|
|
140
|
-
|
|
141
|
-
|
|
142
|
-
|
|
137
|
+
.map((item, idx) => {
|
|
138
|
+
let speaker;
|
|
139
|
+
if (typeof item.speaker === 'string' && item.speaker.trim() !== '') {
|
|
140
|
+
speaker = item.speaker.trim();
|
|
141
|
+
}
|
|
142
|
+
else if (typeof item.speaker === 'number' && Number.isFinite(item.speaker)) {
|
|
143
|
+
speaker = 'Speaker ' + String(item.speaker);
|
|
144
|
+
}
|
|
145
|
+
return {
|
|
146
|
+
id: typeof item.id === 'number' && Number.isFinite(item.id) ? item.id : idx,
|
|
147
|
+
start: typeof item.start === 'number' && Number.isFinite(item.start) ? item.start : 0,
|
|
148
|
+
end: typeof item.end === 'number' && Number.isFinite(item.end) ? item.end : 0,
|
|
149
|
+
text: typeof item.text === 'string' ? item.text.trim() : '',
|
|
150
|
+
...(speaker !== undefined ? { speaker } : {}),
|
|
151
|
+
};
|
|
152
|
+
});
|
|
153
|
+
if (options.diarize === true && segments.length > 0 && segments.every((seg) => seg.speaker === undefined)) {
|
|
154
|
+
let speakerTurn = 0;
|
|
155
|
+
segments[0].speaker = '说话人 A';
|
|
156
|
+
for (let i = 1; i < segments.length; i++) {
|
|
157
|
+
if (segments[i].start - segments[i - 1].end >= 1.2) {
|
|
158
|
+
speakerTurn = 1 - speakerTurn;
|
|
159
|
+
}
|
|
160
|
+
segments[i].speaker = speakerTurn === 0 ? '说话人 A' : '说话人 B';
|
|
161
|
+
}
|
|
162
|
+
}
|
|
143
163
|
}
|
|
164
|
+
const speakerSet = new Set();
|
|
165
|
+
for (const seg of segments ?? []) {
|
|
166
|
+
if (typeof seg.speaker === 'string' && seg.speaker.trim() !== '') {
|
|
167
|
+
speakerSet.add(seg.speaker.trim());
|
|
168
|
+
}
|
|
169
|
+
}
|
|
170
|
+
const speakers = speakerSet.size > 0 ? Array.from(speakerSet) : undefined;
|
|
144
171
|
return {
|
|
145
172
|
text,
|
|
146
173
|
model: options.model,
|
|
147
174
|
...(language !== undefined ? { language } : {}),
|
|
148
175
|
...(duration !== undefined ? { duration } : {}),
|
|
149
176
|
...(segments !== undefined ? { segments } : {}),
|
|
177
|
+
...(speakers !== undefined ? { speakers } : {}),
|
|
150
178
|
};
|
|
151
179
|
}
|
|
152
180
|
function formatSubtitleTimestamp(seconds, decimalSep) {
|
|
@@ -160,6 +188,11 @@ function formatSubtitleTimestamp(seconds, decimalSep) {
|
|
|
160
188
|
const hour = Math.floor(totalMin / 60);
|
|
161
189
|
return `${String(hour).padStart(2, '0')}:${String(min).padStart(2, '0')}:${String(sec).padStart(2, '0')}${decimalSep}${String(ms).padStart(3, '0')}`;
|
|
162
190
|
}
|
|
191
|
+
function formatSegmentText(seg) {
|
|
192
|
+
const clean = seg.text.trim();
|
|
193
|
+
const speaker = typeof seg.speaker === 'string' ? seg.speaker.trim() : '';
|
|
194
|
+
return speaker !== '' ? `[${speaker}] ${clean}` : clean;
|
|
195
|
+
}
|
|
163
196
|
/** 将 ASR 返回的 segments 格式化为 SRT 或 WebVTT 字幕文本。 */
|
|
164
197
|
export function formatSttSubtitles(segments, format) {
|
|
165
198
|
const valid = segments.filter((seg) => seg.text.trim() !== '');
|
|
@@ -167,16 +200,51 @@ export function formatSttSubtitles(segments, format) {
|
|
|
167
200
|
if (valid.length === 0)
|
|
168
201
|
return 'WEBVTT\n';
|
|
169
202
|
const body = valid
|
|
170
|
-
.map((seg) => `${formatSubtitleTimestamp(seg.start, '.')} --> ${formatSubtitleTimestamp(seg.end, '.')}\n${seg
|
|
203
|
+
.map((seg) => `${formatSubtitleTimestamp(seg.start, '.')} --> ${formatSubtitleTimestamp(seg.end, '.')}\n${formatSegmentText(seg)}`)
|
|
171
204
|
.join('\n\n');
|
|
172
205
|
return `WEBVTT\n\n${body}\n`;
|
|
173
206
|
}
|
|
174
207
|
if (valid.length === 0)
|
|
175
208
|
return '';
|
|
176
209
|
return (valid
|
|
177
|
-
.map((seg, idx) => `${idx + 1}\n${formatSubtitleTimestamp(seg.start, ',')} --> ${formatSubtitleTimestamp(seg.end, ',')}\n${seg
|
|
210
|
+
.map((seg, idx) => `${idx + 1}\n${formatSubtitleTimestamp(seg.start, ',')} --> ${formatSubtitleTimestamp(seg.end, ',')}\n${formatSegmentText(seg)}`)
|
|
178
211
|
.join('\n\n') + '\n');
|
|
179
212
|
}
|
|
213
|
+
function formatTranscriptTimestamp(seconds) {
|
|
214
|
+
const safe = Number.isFinite(seconds) ? Math.max(0, Math.floor(seconds)) : 0;
|
|
215
|
+
const hours = Math.floor(safe / 3600);
|
|
216
|
+
const minutes = Math.floor((safe % 3600) / 60);
|
|
217
|
+
const secs = safe % 60;
|
|
218
|
+
if (hours > 0) {
|
|
219
|
+
return `${String(hours).padStart(2, '0')}:${String(minutes).padStart(2, '0')}:${String(secs).padStart(2, '0')}`;
|
|
220
|
+
}
|
|
221
|
+
return `${String(minutes).padStart(2, '0')}:${String(secs).padStart(2, '0')}`;
|
|
222
|
+
}
|
|
223
|
+
/** 将带说话人标签的分段合并为按发言人分段的可读逐字稿(如 `[说话人 A 00:01] ...`)。 */
|
|
224
|
+
export function formatDiarizedTranscript(segments) {
|
|
225
|
+
const valid = segments.filter((seg) => seg.text.trim() !== '');
|
|
226
|
+
if (valid.length === 0)
|
|
227
|
+
return '';
|
|
228
|
+
const groups = [];
|
|
229
|
+
for (const seg of valid) {
|
|
230
|
+
const speaker = typeof seg.speaker === 'string' && seg.speaker.trim() !== ''
|
|
231
|
+
? seg.speaker.trim()
|
|
232
|
+
: '说话人';
|
|
233
|
+
const part = seg.text.trim();
|
|
234
|
+
const last = groups[groups.length - 1];
|
|
235
|
+
if (last && last.speaker === speaker) {
|
|
236
|
+
const needsSpace = !/[\u4e00-\u9fff\u3000-\u303f\uff00-\uffef]$/.test(last.text)
|
|
237
|
+
|| !/^[\u4e00-\u9fff]/.test(part);
|
|
238
|
+
last.text += (needsSpace ? ' ' : '') + part;
|
|
239
|
+
}
|
|
240
|
+
else {
|
|
241
|
+
groups.push({ speaker, start: seg.start, text: part });
|
|
242
|
+
}
|
|
243
|
+
}
|
|
244
|
+
return groups
|
|
245
|
+
.map((group) => `[${group.speaker} ${formatTranscriptTimestamp(group.start)}] ${group.text}`)
|
|
246
|
+
.join('\n\n');
|
|
247
|
+
}
|
|
180
248
|
/** 检测环境 PATH 中是否可用 ffmpeg。 */
|
|
181
249
|
export function hasFfmpeg(env = process.env) {
|
|
182
250
|
try {
|
|
@@ -258,6 +326,86 @@ export async function prepareAudioForAsr(audioPath, signal) {
|
|
|
258
326
|
throw error;
|
|
259
327
|
}
|
|
260
328
|
}
|
|
329
|
+
async function probeAudioChannels(audioPath, signal) {
|
|
330
|
+
throwIfAborted(signal);
|
|
331
|
+
const viaFfprobe = await new Promise((resolve) => {
|
|
332
|
+
execFile('ffprobe', ['-v', 'error', '-select_streams', 'a:0', '-show_entries', 'stream=channels', '-of', 'csv=p=0', audioPath], { signal, maxBuffer: 1024 * 1024 }, (error, stdout) => {
|
|
333
|
+
if (error) {
|
|
334
|
+
resolve(null);
|
|
335
|
+
return;
|
|
336
|
+
}
|
|
337
|
+
const parsed = Number.parseInt(String(stdout).trim(), 10);
|
|
338
|
+
resolve(Number.isFinite(parsed) && parsed > 0 ? parsed : null);
|
|
339
|
+
});
|
|
340
|
+
});
|
|
341
|
+
throwIfAborted(signal);
|
|
342
|
+
if (viaFfprobe !== null)
|
|
343
|
+
return viaFfprobe;
|
|
344
|
+
return await new Promise((resolve) => {
|
|
345
|
+
execFile('ffmpeg', ['-i', audioPath], { signal, maxBuffer: 2 * 1024 * 1024 }, (_error, _stdout, stderr) => {
|
|
346
|
+
const text = typeof stderr === 'string' ? stderr : '';
|
|
347
|
+
const audioLine = text.split(/\r?\n/).find((line) => /\bAudio:\b/.test(line)) ?? '';
|
|
348
|
+
if (/\b(?:stereo|quad|5\.1|7\.1)\b/i.test(audioLine) || /\b2 channels\b/i.test(audioLine)) {
|
|
349
|
+
resolve(2);
|
|
350
|
+
return;
|
|
351
|
+
}
|
|
352
|
+
const chMatch = /\b(\d+)\s+channels?\b/i.exec(audioLine);
|
|
353
|
+
if (chMatch) {
|
|
354
|
+
resolve(Number.parseInt(chMatch[1], 10) || 1);
|
|
355
|
+
return;
|
|
356
|
+
}
|
|
357
|
+
resolve(1);
|
|
358
|
+
});
|
|
359
|
+
});
|
|
360
|
+
}
|
|
361
|
+
/**
|
|
362
|
+
* 探测音频是否包含 >= 2 声道;若为立体声/多声道,拆分为左右声道(16kHz 32kbps 单声道 MP3)用于精准分轨说话人分离。
|
|
363
|
+
* 若非立体声或 ffmpeg 不可用/失败,返回 null。
|
|
364
|
+
*/
|
|
365
|
+
export async function splitStereoChannelsForDiarization(audioPath, signal) {
|
|
366
|
+
throwIfAborted(signal);
|
|
367
|
+
let channels = 1;
|
|
368
|
+
try {
|
|
369
|
+
channels = await probeAudioChannels(audioPath, signal);
|
|
370
|
+
}
|
|
371
|
+
catch {
|
|
372
|
+
throwIfAborted(signal);
|
|
373
|
+
return null;
|
|
374
|
+
}
|
|
375
|
+
throwIfAborted(signal);
|
|
376
|
+
if (channels < 2)
|
|
377
|
+
return null;
|
|
378
|
+
const tempDir = mkdtempSync(join(tmpdir(), 'dsh-voice-stereo-'));
|
|
379
|
+
const cleanup = () => {
|
|
380
|
+
try {
|
|
381
|
+
rmSync(tempDir, { recursive: true, force: true });
|
|
382
|
+
}
|
|
383
|
+
catch {
|
|
384
|
+
// 尽力清理临时目录
|
|
385
|
+
}
|
|
386
|
+
};
|
|
387
|
+
try {
|
|
388
|
+
const leftPath = join(tempDir, 'channel_left.mp3');
|
|
389
|
+
const rightPath = join(tempDir, 'channel_right.mp3');
|
|
390
|
+
await runFfmpeg(['-y', '-i', audioPath, '-vn', '-af', 'pan=mono|c0=c0', '-ar', '16000', '-ac', '1', '-b:a', '32k', leftPath], signal);
|
|
391
|
+
await runFfmpeg(['-y', '-i', audioPath, '-vn', '-af', 'pan=mono|c0=c1', '-ar', '16000', '-ac', '1', '-b:a', '32k', rightPath], signal);
|
|
392
|
+
throwIfAborted(signal);
|
|
393
|
+
const leftBlob = await openAsBlob(leftPath, { type: 'audio/mpeg' });
|
|
394
|
+
const rightBlob = await openAsBlob(rightPath, { type: 'audio/mpeg' });
|
|
395
|
+
return {
|
|
396
|
+
channels: [
|
|
397
|
+
{ audio: leftBlob, filename: 'channel_left.mp3', speaker: '说话人 A' },
|
|
398
|
+
{ audio: rightBlob, filename: 'channel_right.mp3', speaker: '说话人 B' },
|
|
399
|
+
],
|
|
400
|
+
cleanup,
|
|
401
|
+
};
|
|
402
|
+
}
|
|
403
|
+
catch {
|
|
404
|
+
cleanup();
|
|
405
|
+
throwIfAborted(signal);
|
|
406
|
+
return null;
|
|
407
|
+
}
|
|
408
|
+
}
|
|
261
409
|
/** 常见音频扩展名 → MIME。 */
|
|
262
410
|
function mimeOf(filename) {
|
|
263
411
|
const ext = extname(filename).toLowerCase();
|
package/lib/tools.d.ts
CHANGED
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import { type ResolvedVoiceConfig } from './config.js';
|
|
2
2
|
import { fetchEdgeVoices, synthesizeSpeech, synthesizeSpeechWithMetadata } from './edge-tts.js';
|
|
3
3
|
import { synthesizeOpenAiSpeech } from './openai-tts.js';
|
|
4
|
-
import { prepareAudioForAsr, transcribe } from './stt.js';
|
|
5
|
-
import { normalizeAudioLoudness } from './tts-text.js';
|
|
4
|
+
import { prepareAudioForAsr, splitStereoChannelsForDiarization, transcribe } from './stt.js';
|
|
5
|
+
import { mixAudioWithBgm, normalizeAudioLoudness } from './tts-text.js';
|
|
6
6
|
/** 模型可见的内容块。 */
|
|
7
7
|
export interface ContentBlock {
|
|
8
8
|
type: 'text';
|
|
@@ -36,6 +36,8 @@ export interface VoiceToolDeps {
|
|
|
36
36
|
prepareAudio?: typeof prepareAudioForAsr;
|
|
37
37
|
fetchVoices?: typeof fetchEdgeVoices;
|
|
38
38
|
normalizeAudio?: typeof normalizeAudioLoudness;
|
|
39
|
+
mixBgm?: typeof mixAudioWithBgm;
|
|
40
|
+
splitStereo?: typeof splitStereoChannelsForDiarization;
|
|
39
41
|
}
|
|
40
42
|
/**
|
|
41
43
|
* 构建五个语音工具定义(支持静态配置对象或活引用配置 getter)。
|