dsh-podwave 0.10.15 → 0.10.20

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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "dsh-podwave",
3
- "version": "0.10.15",
3
+ "version": "0.10.20",
4
4
  "description": "DeepSeek Harness 插件:本地优先的播客播放器 + 转写 + 双语字幕 + 字幕翻译。",
5
5
  "type": "module",
6
6
  "main": "./src/host.js",
@@ -23,7 +23,7 @@
23
23
  },
24
24
  "engines": {
25
25
  "node": ">=20",
26
- "dsh": ">=0.2.1-rc.5 <0.3.0"
26
+ "dsh": ">=0.2.1-rc.5 <0.3.0 || >=0.2.2-rc.0 <0.3.0 || >=0.2.3-rc.0 <0.3.0 || >=0.2.4-rc.0 <0.3.0"
27
27
  },
28
28
  "keywords": [
29
29
  "dsh",
@@ -16,7 +16,10 @@ def main():
16
16
  parser.add_argument("--model-dir", required=True)
17
17
  parser.add_argument("--out-dir", required=True)
18
18
  parser.add_argument("--threads", type=int, default=4)
19
+ parser.add_argument("--hotwords", default="")
19
20
  args = parser.parse_args()
21
+ if args.hotwords.strip():
22
+ print("note: SenseVoice 模型不支持热词,已忽略 --hotwords", file=sys.stderr)
20
23
  os.makedirs(args.out_dir, exist_ok=True)
21
24
  for wav in args.files:
22
25
  base = os.path.splitext(os.path.basename(wav))[0]
package/src/downloader.js CHANGED
@@ -4,6 +4,7 @@ import fs from 'node:fs'
4
4
  import fsp from 'node:fs/promises'
5
5
  import path from 'node:path'
6
6
  import { spawn } from 'node:child_process'
7
+ import crypto from 'node:crypto'
7
8
  import { Readable } from 'node:stream'
8
9
  import { pipeline } from 'node:stream/promises'
9
10
 
@@ -38,7 +39,7 @@ export async function listHfRepo(repo, base, { include, skip } = {}) {
38
39
  .filter((entry) => entry.type === 'file')
39
40
  .filter((entry) => (include ? include.includes(entry.path) : true))
40
41
  .filter((entry) => (skip ? !skip.some((re) => re.test(entry.path)) : true))
41
- .map((entry) => ({ path: entry.path, size: entry.size || 0 }))
42
+ .map((entry) => ({ path: entry.path, size: entry.size || 0, sha256: entry.lfs?.oid || '' }))
42
43
  }
43
44
 
44
45
  export async function directHfFiles(repo, base, { include, files, skip } = {}) {
@@ -74,11 +75,93 @@ export async function listMsRepo(repo) {
74
75
  .map((file) => ({ path: file.Path, size: file.Size || 0 }))
75
76
  }
76
77
 
77
- async function downloadOne(url, dest, { expected = 0, onProgress, signal } = {}) {
78
+ export function isSafeArchivePath(name) {
79
+ const value = String(name || '').replace(/\\/g, '/')
80
+ if (!value || value.startsWith('/')) return false
81
+ if (/^[a-zA-Z]:/.test(value)) return false
82
+ return !value.split('/').some((part) => part === '..')
83
+ }
84
+
85
+ function runCapture(cmd, args) {
86
+ return new Promise((resolve) => {
87
+ let output = ''
88
+ const child = spawn(cmd, args, { stdio: ['ignore', 'pipe', 'ignore'] })
89
+ child.stdout.on('data', (chunk) => {
90
+ output += chunk.toString('utf8')
91
+ })
92
+ child.on('error', () => resolve(null))
93
+ child.on('exit', (code) => resolve(code === 0 ? output : null))
94
+ })
95
+ }
96
+
97
+ async function listArchiveEntries(file) {
98
+ const lower = file.toLowerCase()
99
+ if (lower.endsWith('.zip')) {
100
+ const candidates = [
101
+ ['unzip', ['-Z1', file]],
102
+ ['tar', ['-tf', file]],
103
+ ['python3', ['-c', 'import zipfile,sys;print("\n".join(zipfile.ZipFile(sys.argv[1]).namelist()))', file]],
104
+ ]
105
+ for (const [cmd, args] of candidates) {
106
+ const output = await runCapture(cmd, args)
107
+ if (output) return output.split(/\r?\n/).filter(Boolean)
108
+ }
109
+ return null
110
+ }
111
+ const output = await runCapture('tar', ['-tf', file])
112
+ return output ? output.split(/\r?\n/).filter(Boolean) : null
113
+ }
114
+
115
+ async function sha256File(file) {
116
+ const hash = crypto.createHash('sha256')
117
+ await pipeline(fs.createReadStream(file), hash)
118
+ return hash.digest('hex')
119
+ }
120
+
121
+ const digestCache = new Map()
122
+
123
+ async function githubDigest(url) {
124
+ try {
125
+ const match = /^https:\/\/github\.com\/([^/]+)\/([^/]+)\/releases\/download\/([^/]+)\/(.+)$/.exec(url)
126
+ if (!match) return ''
127
+ const [, owner, repo, tag, assetName] = match
128
+ const key = `${owner}/${repo}@${tag}#${assetName}`
129
+ if (digestCache.has(key)) return digestCache.get(key)
130
+ const api =
131
+ tag === 'latest'
132
+ ? `https://api.github.com/repos/${owner}/${repo}/releases/latest`
133
+ : `https://api.github.com/repos/${owner}/${repo}/releases/tags/${tag}`
134
+ const res = await fetch(api, {
135
+ headers: { 'user-agent': UA, accept: 'application/vnd.github+json' },
136
+ signal: timeout(15000),
137
+ })
138
+ if (!res.ok) {
139
+ digestCache.set(key, '')
140
+ return ''
141
+ }
142
+ const data = await res.json()
143
+ const asset = (data.assets || []).find((item) => item.name === decodeURIComponent(assetName))
144
+ const digest = asset?.digest ? String(asset.digest).replace(/^sha256:/, '') : ''
145
+ digestCache.set(key, digest)
146
+ return digest
147
+ } catch {
148
+ return ''
149
+ }
150
+ }
151
+
152
+ async function downloadOne(url, dest, { expected = 0, sha256 = '', onProgress, signal } = {}) {
78
153
  await fsp.mkdir(path.dirname(dest), { recursive: true })
79
154
  try {
80
155
  const stat = await fsp.stat(dest)
81
- if (expected && stat.size === expected) return { bytes: stat.size, skipped: true }
156
+ if (expected && stat.size === expected) {
157
+ if (sha256) {
158
+ const actual = await sha256File(dest)
159
+ if (actual.toLowerCase() === String(sha256).toLowerCase()) return { bytes: stat.size, skipped: true }
160
+ await fsp.rm(dest, { force: true })
161
+ } else {
162
+ return { bytes: stat.size, skipped: true }
163
+ }
164
+ }
82
165
  if (expected && stat.size > expected) await fsp.rm(dest, { force: true })
83
166
  } catch {
84
167
  // 文件不存在
@@ -108,6 +191,13 @@ async function downloadOne(url, dest, { expected = 0, onProgress, signal } = {})
108
191
  await pipeline(source, write)
109
192
  const stat = await fsp.stat(part)
110
193
  if (expected && stat.size !== expected) throw new Error(`文件大小不符(期望 ${expected},实际 ${stat.size})`)
194
+ if (sha256) {
195
+ const actual = await sha256File(part)
196
+ if (actual.toLowerCase() !== String(sha256).toLowerCase()) {
197
+ await fsp.rm(part, { force: true })
198
+ throw new Error('文件校验失败(sha256 不匹配),已丢弃下载内容')
199
+ }
200
+ }
111
201
  await fsp.rename(part, dest)
112
202
  return { bytes: stat.size, skipped: false }
113
203
  }
@@ -161,7 +251,8 @@ export async function downloadArtifact(artifact, { sourceId = 'auto', onProgress
161
251
  const url = source.id === 'cn' ? `https://gh-proxy.com/${asset.url}` : asset.url
162
252
  const dest = path.join(destDir, asset.file)
163
253
  const before = result.bytes
164
- const { bytes } = await downloadOne(url, dest, { onProgress, signal })
254
+ const sha256 = await githubDigest(asset.url)
255
+ const { bytes } = await downloadOne(url, dest, { sha256, onProgress, signal })
165
256
  result.bytes += bytes
166
257
  result.files.push({ path: asset.file, bytes })
167
258
  if (asset.extract) await extractArchive(dest, destDir)
@@ -214,6 +305,7 @@ export async function downloadArtifact(artifact, { sourceId = 'auto', onProgress
214
305
  const dest = path.join(destDir, file.path)
215
306
  const { bytes } = await downloadOne(url, dest, {
216
307
  expected: file.size,
308
+ sha256: file.sha256 || '',
217
309
  signal,
218
310
  onProgress: (event) => {
219
311
  done += event.delta
@@ -239,6 +331,11 @@ function runExtract(cmd, args) {
239
331
  }
240
332
 
241
333
  export async function extractArchive(file, destDir) {
334
+ const entries = await listArchiveEntries(file)
335
+ if (entries && entries.length) {
336
+ const unsafe = entries.find((entry) => !isSafeArchivePath(entry))
337
+ if (unsafe) throw new Error(`压缩包包含不安全路径:${unsafe}`)
338
+ }
242
339
  const lower = file.toLowerCase()
243
340
  const attempts = []
244
341
  if (lower.endsWith('.zip')) {
package/src/host.js CHANGED
@@ -8,7 +8,7 @@ import path from 'node:path'
8
8
  import { createReadStream, createWriteStream } from 'node:fs'
9
9
  import { execFile } from 'node:child_process'
10
10
  import { pipeline } from 'node:stream/promises'
11
- import { DATA_DIR } from './state.js'
11
+ import { DATA_DIR, COVERS_DIR } from './state.js'
12
12
  import {
13
13
  addFolderState,
14
14
  getState,
@@ -142,7 +142,9 @@ function boot() {
142
142
  await initState()
143
143
  await migrateAllTranscripts().catch((error) => console.error('[dsh-podwave] transcript migration failed:', error.message))
144
144
  if (getState().folders.length) {
145
- scanLibrary().catch((error) => console.error('[dsh-podwave] initial scan failed:', error.message))
145
+ scanLibrary()
146
+ .then(() => refreshWatchers())
147
+ .catch((error) => console.error('[dsh-podwave] initial scan failed:', error.message))
146
148
  }
147
149
  })().catch((error) => {
148
150
  booted = null
@@ -386,7 +388,7 @@ async function route(ctx, req, res) {
386
388
  }
387
389
  if ((method === 'GET' || method === 'HEAD') && routePath === 'api/health') {
388
390
  await mediaReady
389
- return json(res, 200, { ok: true, app: 'podwave', version: '0.10.15', mediaBase, transcribe: runtimeInfo(), diarize: diarizeInfo(), translate: publicConfig() })
391
+ return json(res, 200, { ok: true, app: 'podwave', version: '0.10.20', mediaBase, transcribe: runtimeInfo(), diarize: diarizeInfo(), translate: publicConfig() })
390
392
  }
391
393
  if (method === 'GET' && routePath === 'api/state') {
392
394
  const state = getState()
@@ -508,8 +510,44 @@ async function route(ctx, req, res) {
508
510
  const id = decodeURIComponent(hideMatch[1])
509
511
  const hidden = new Set(getState().hidden || [])
510
512
  hidden.add(id)
511
- updateState({ hidden: [...hidden] })
512
- return json(res, 200, { id, hidden: true })
513
+ await deleteTranscript(id).catch(() => {})
514
+ for (const ext of ['jpg', 'png']) await rm(path.join(COVERS_DIR, `${id}.${ext}`), { force: true }).catch(() => {})
515
+ const favorites = new Set(getState().favorites || [])
516
+ favorites.delete(id)
517
+ const progress = { ...(getState().progress || {}) }
518
+ delete progress[id]
519
+ updateState({ hidden: [...hidden], favorites: [...favorites], progress })
520
+ return json(res, 200, { id, hidden: true, purged: true })
521
+ }
522
+
523
+ if (routePath === 'api/search' && method === 'GET') {
524
+ const query = String(new URL(req.url, 'http://127.0.0.1').searchParams.get('q') || '').trim()
525
+ if (query.length < 2) return json(res, 200, { items: [], total: 0 })
526
+ const needle = query.toLowerCase()
527
+ const titles = new Map(getEpisodes().map((episode) => [episode.id, episode]))
528
+ const items = []
529
+ const names = await readdir(TRANSCRIPTS_DIR).catch(() => [])
530
+ for (const name of names) {
531
+ if (!name.endsWith('.json') || items.length >= 80) continue
532
+ const id = name.slice(0, -5)
533
+ const transcript = await readTranscript(id).catch(() => null)
534
+ if (!transcript) continue
535
+ for (const line of transcript.lines || []) {
536
+ if (items.length >= 80) break
537
+ if (!String(line.text || '').toLowerCase().includes(needle)) continue
538
+ const episode = titles.get(id)
539
+ items.push({
540
+ episodeId: id,
541
+ lineId: line.id,
542
+ start: line.start,
543
+ end: line.end,
544
+ text: line.text,
545
+ title: episode?.title || transcript.title || '',
546
+ episodeTitle: episode?.title || transcript.title || '',
547
+ })
548
+ }
549
+ }
550
+ return json(res, 200, { items, total: items.length })
513
551
  }
514
552
 
515
553
  if (routePath === 'api/import/settings') {
@@ -542,7 +580,9 @@ async function route(ctx, req, res) {
542
580
  const body = await readJson(req)
543
581
  const folder = await addFolder(body?.path)
544
582
  addFolderState(folder)
545
- scanLibrary().catch((error) => console.error('[dsh-podwave] scan failed:', error.message))
583
+ scanLibrary()
584
+ .then(() => refreshWatchers())
585
+ .catch((error) => console.error('[dsh-podwave] scan failed:', error.message))
546
586
  return json(res, 200, { folder, folders: getState().folders })
547
587
  }
548
588
  if (method === 'DELETE') {
@@ -556,7 +596,9 @@ async function route(ctx, req, res) {
556
596
  const removed = forgetEpisodesUnder(target)
557
597
  removeProgressFor((id) => removed.has(id))
558
598
  removeFolderState(target)
559
- scanLibrary().catch((error) => console.error('[dsh-podwave] scan failed:', error.message))
599
+ scanLibrary()
600
+ .then(() => refreshWatchers())
601
+ .catch((error) => console.error('[dsh-podwave] scan failed:', error.message))
560
602
  return json(res, 200, { folders: next })
561
603
  }
562
604
  return json(res, 405, { error: '不支持的方法' })
package/src/setup.js CHANGED
@@ -9,7 +9,7 @@ import { getSection, updateSection } from './config.js'
9
9
  import { getJob, upsertJob } from './jobs.js'
10
10
  import { CATALOG, FFMPEG_BINARY, LLAMA_BINARY, SOURCE_LABELS, artifactFor, ffmpegAsset, llamaAsset, sourcesFor } from './catalog.js'
11
11
  import { downloadArtifact, extractArchive } from './downloader.js'
12
- import { detectPlatform, findExecutable, ffmpegPath, ffprobePath, recommendEngine, venvDir, venvPython, isWindows } from './platform.js'
12
+ import { detectPlatform, findExecutable, ffmpegPath, ffprobePath, recommendEngine, venvDir, venvPython, isWindows, isMac } from './platform.js'
13
13
 
14
14
  const RUNTIME_DIR = path.join(DATA_DIR, 'runtime')
15
15
  const SETUP_PATH = path.join(DATA_DIR, 'setup.json')
@@ -338,7 +338,7 @@ export function scan() {
338
338
  label: 'ffmpeg / ffprobe',
339
339
  required: true,
340
340
  ready: Boolean(ffmpegPath() && ffprobePath()),
341
- detail: ffmpegPath() || (ffAsset ? '未找到,将下载静态构建' : '未找到;macOS 请先 brew install ffmpeg'),
341
+ detail: ffmpegPath() || (ffAsset ? '未找到,将下载静态构建' : isMac && findExecutable('brew') ? '未找到,一键安装将通过 Homebrew 安装' : '未找到;macOS 请先 brew install ffmpeg'),
342
342
  bytes: ffAsset?.bytes || 0,
343
343
  })
344
344
  push({
@@ -427,9 +427,18 @@ export async function install({ source = 'auto' } = {}) {
427
427
  // 0. ffmpeg(Windows / Linux 直接下载静态构建;macOS 优先用系统/Homebrew)
428
428
  if (!ffmpegPath()) {
429
429
  const ffAsset = ffmpegAsset(platform)
430
- if (!ffAsset) {
431
- throw new Error('没有找到 ffmpeg:macOS 请先运行 brew install ffmpeg,或把 ffmpeg/ffprobe 放进 PATH')
432
- }
430
+ if (isMac) {
431
+ const brew = findExecutable('brew')
432
+ if (!brew) throw new Error('没有找到 ffmpeg,也没有 Homebrew;请安装 Homebrew 或把 ffmpeg/ffprobe 放进 PATH')
433
+ report({ stage: '通过 Homebrew 安装 ffmpeg' })
434
+ await run(brew, ['install', 'ffmpeg'])
435
+ if (!ffmpegPath()) throw new Error('Homebrew 安装后仍未找到 ffmpeg,请检查 brew 输出')
436
+ summary.push({ id: 'ffmpeg', label: 'ffmpeg / ffprobe(Homebrew)', bytes: 0 })
437
+ doneBytes += 0
438
+ report({ stage: '检查环境' })
439
+ } else if (!ffAsset) {
440
+ throw new Error('没有找到 ffmpeg:请安装 ffmpeg 或把它放进 PATH')
441
+ } else {
433
442
  report({ stage: '下载 ffmpeg' })
434
443
  await downloadArtifact(
435
444
  {
@@ -446,6 +455,7 @@ export async function install({ source = 'auto' } = {}) {
446
455
  if (fs.existsSync(archive)) await extractArchive(archive, target)
447
456
  doneBytes += ffAsset.bytes
448
457
  summary.push({ id: 'ffmpeg', label: 'ffmpeg / ffprobe', bytes: ffAsset.bytes })
458
+ }
449
459
  }
450
460
 
451
461
  // 1. Python 运行时
package/src/transcribe.js CHANGED
@@ -225,6 +225,24 @@ function waitChild(child) {
225
225
  })
226
226
  }
227
227
 
228
+ function runCapture(command, args) {
229
+ return new Promise((resolve) => {
230
+ let output = ''
231
+ let child
232
+ try {
233
+ child = spawn(command, args, { stdio: ['ignore', 'ignore', 'pipe'] })
234
+ } catch {
235
+ resolve('')
236
+ return
237
+ }
238
+ child.stderr.on('data', (chunk) => {
239
+ output += chunk.toString('utf8')
240
+ })
241
+ child.on('error', () => resolve(''))
242
+ child.on('close', () => resolve(output))
243
+ })
244
+ }
245
+
228
246
  function runProcess(command, args) {
229
247
  return new Promise((resolve) => {
230
248
  let child
@@ -260,6 +278,64 @@ async function probeDuration(file) {
260
278
  }
261
279
 
262
280
  // 用 ffmpeg 把长音频切成若干块;先尝试不重编码(-c copy),失败再转 AAC。
281
+ async function detectSilenceMidpoints(file) {
282
+ const output = await runCapture(FFMPEG, [
283
+ '-hide_banner',
284
+ '-nostats',
285
+ '-i', file,
286
+ '-af', 'silencedetect=noise=-30dB:d=0.45',
287
+ '-f', 'null',
288
+ '-',
289
+ ])
290
+ const starts = []
291
+ const mids = []
292
+ for (const line of output.split(/\r?\n/)) {
293
+ const start = /silence_start: (-?[\d.]+)/.exec(line)
294
+ if (start) starts.push(Number(start[1]))
295
+ const end = /silence_end: ([\d.]+)/.exec(line)
296
+ if (end && starts.length) {
297
+ const from = starts.pop()
298
+ const to = Number(end[1])
299
+ if (to > from) mids.push((from + to) / 2)
300
+ }
301
+ }
302
+ return mids
303
+ }
304
+
305
+ function planCuts(duration, chunkSeconds, silences) {
306
+ const cuts = [0]
307
+ let target = chunkSeconds
308
+ while (target < duration - 20) {
309
+ const floor = cuts[cuts.length - 1] + Math.min(chunkSeconds * 0.5, 30)
310
+ const window = chunkSeconds * 0.45
311
+ const candidates = silences.filter((time) => time > floor && time < duration - 8 && Math.abs(time - target) <= window)
312
+ const pick = candidates.sort((a, b) => Math.abs(a - target) - Math.abs(b - target))[0]
313
+ cuts.push(round2(pick ?? target))
314
+ target = (pick ?? target) + chunkSeconds
315
+ }
316
+ cuts.push(round2(duration))
317
+ return cuts.filter((value, index, list) => index === 0 || value > list[index - 1] + 5)
318
+ }
319
+
320
+ async function splitAudioAt(file, dir, cuts) {
321
+ await mkdir(dir, { recursive: true })
322
+ const sherpa = currentAsrModel().engine === 'sherpa'
323
+ const parts = []
324
+ for (let index = 0; index < cuts.length - 1; index += 1) {
325
+ const start = cuts[index]
326
+ const end = cuts[index + 1]
327
+ const ext = sherpa ? 'wav' : 'm4a'
328
+ const out = path.join(dir, `chunk-${String(index).padStart(3, '0')}.${ext}`)
329
+ const code = sherpa
330
+ ? await runProcess(FFMPEG, ['-hide_banner', '-loglevel', 'error', '-y', '-i', file, '-ss', String(start), '-to', String(end), '-ar', '16000', '-ac', '1', '-c:a', 'pcm_s16le', out])
331
+ : await runProcess(FFMPEG, ['-hide_banner', '-loglevel', 'error', '-y', '-i', file, '-ss', String(start), '-to', String(end), '-vn', '-c:a', 'aac', '-b:a', '96k', out])
332
+ if (code !== 0) throw new Error('音频切分失败(ffmpeg)')
333
+ const actual = (await probeDuration(out)) || end - start
334
+ parts.push({ index, start: round2(start), end: round2(start + actual), file: out, duration: round2(actual) })
335
+ }
336
+ return parts
337
+ }
338
+
263
339
  async function splitAudio(file, dir, chunkSeconds) {
264
340
  await mkdir(dir, { recursive: true })
265
341
  const sherpa = currentAsrModel().engine === 'sherpa'
@@ -312,9 +388,16 @@ async function ensurePlan(episode, workDir) {
312
388
  await rm(path.join(workDir, 'parts'), { recursive: true, force: true })
313
389
  await rm(path.join(workDir, 'results'), { recursive: true, force: true })
314
390
  const single = currentAsrModel().engine !== 'sherpa' && duration <= chunkSeconds * 1.2
315
- const parts = single
316
- ? [{ index: 0, start: 0, end: round2(duration), file: episode.file, duration: round2(duration) }]
317
- : await splitAudio(episode.file, path.join(workDir, 'parts'), chunkSeconds)
391
+ let parts
392
+ if (single) {
393
+ parts = [{ index: 0, start: 0, end: round2(duration), file: episode.file, duration: round2(duration) }]
394
+ } else {
395
+ const silences = await detectSilenceMidpoints(episode.file).catch(() => [])
396
+ const cuts = planCuts(duration, chunkSeconds, silences)
397
+ parts = cuts.length > 2
398
+ ? await splitAudioAt(episode.file, path.join(workDir, 'parts'), cuts)
399
+ : await splitAudio(episode.file, path.join(workDir, 'parts'), chunkSeconds)
400
+ }
318
401
  const plan = { duration: round2(duration), chunkSeconds, chunked: !single, parts }
319
402
  await writeFile(metaPath, JSON.stringify(plan, null, 2), 'utf8')
320
403
  return plan
@@ -370,10 +453,12 @@ function wordsOf(raw) {
370
453
 
371
454
  function buildBatchArgs(files, outDir) {
372
455
  const model = currentAsrModel()
456
+ const { hotwords } = transcribeConfig()
373
457
  if (model.engine === 'sherpa') {
374
- return [SHERPA_SCRIPT, ...files, '--model-dir', model.dir, '--out-dir', outDir, '--threads', '4']
458
+ const args = [SHERPA_SCRIPT, ...files, '--model-dir', model.dir, '--out-dir', outDir, '--threads', '4']
459
+ if (hotwords.length) args.push('--hotwords', hotwords.join(' '))
460
+ return args
375
461
  }
376
- const { hotwords } = transcribeConfig()
377
462
  const args = [
378
463
  ...files,
379
464
  '--model', currentAsrModel().dir,
@@ -703,9 +788,9 @@ function groupWords(words) {
703
788
  const elapsed = current.end - current.start
704
789
  if (
705
790
  /[。!?!?…]$/.test(joined) ||
706
- (/[,,;;::]$/.test(joined) && joined.length > 28) ||
707
- joined.length >= 60 ||
708
- elapsed >= 12
791
+ (/[,,;;::]$/.test(joined) && joined.length > 42) ||
792
+ joined.length >= 90 ||
793
+ elapsed >= 15
709
794
  ) {
710
795
  flush()
711
796
  }
@@ -812,18 +897,21 @@ function buildSegments(raw, words) {
812
897
  cursor += 1
813
898
  }
814
899
  appendWord(token.text, matched ? matched.start : chunkStart, matched ? matched.end : chunkEnd)
815
- if (current && (current.text.length >= 60 || current.end - current.start >= 12)) flush()
900
+ if (current && /[,,;;]$/.test(current.text) && current.text.length >= 48) flush()
901
+ if (current && (current.text.length >= 90 || current.end - current.start >= 15)) flush()
816
902
  }
817
- flush()
903
+ if (current && /[。!?!?…]$/.test(current.text) && current.text.length > 2) flush()
818
904
  }
819
905
  flush()
820
906
 
821
- return lines.map((line, index) => ({
822
- id: index + 1,
823
- start: round2(line.start),
824
- end: round2(line.end),
825
- text: line.text.trim(),
826
- }))
907
+ return lines
908
+ .filter((line) => line.text.trim() && line.end - line.start >= 0.2)
909
+ .map((line, index) => ({
910
+ id: index + 1,
911
+ start: round2(line.start),
912
+ end: round2(line.end),
913
+ text: line.text.trim(),
914
+ }))
827
915
  }
828
916
 
829
917
  export async function startTranscription(episode, { force = false, resume = false } = {}) {
package/src/translate.js CHANGED
@@ -70,6 +70,34 @@ function parseNumbered(text, count) {
70
70
  return result
71
71
  }
72
72
 
73
+ async function fetchWithRetry(url, init, { attempts = 3, timeoutMs = 120000 } = {}) {
74
+ let lastError = null
75
+ for (let attempt = 1; attempt <= attempts; attempt += 1) {
76
+ try {
77
+ const response = await fetch(url, { ...init, signal: AbortSignal.timeout(timeoutMs) })
78
+ if (response.status === 429 || response.status >= 500) {
79
+ lastError = new Error(`翻译 API ${response.status}`)
80
+ if (attempt < attempts) {
81
+ await new Promise((resolve) => setTimeout(resolve, attempt * 1500))
82
+ continue
83
+ }
84
+ const detail = await response.text().catch(() => '')
85
+ throw new Error(`翻译 API ${response.status}:${detail.slice(0, 200)}`)
86
+ }
87
+ return response
88
+ } catch (error) {
89
+ lastError = error
90
+ const retryable = error?.name === 'TimeoutError' || /fetch failed|network|ECONN|ETIMEDOUT/i.test(String(error?.message || error))
91
+ if (retryable && attempt < attempts) {
92
+ await new Promise((resolve) => setTimeout(resolve, attempt * 1500))
93
+ continue
94
+ }
95
+ throw error
96
+ }
97
+ }
98
+ throw lastError
99
+ }
100
+
73
101
  async function translateWithApi(items, target, config) {
74
102
  const language = target === 'zh' ? 'Simplified Chinese' : 'English'
75
103
  const system = `You are a professional subtitle translator. Translate each numbered line into ${language}. Preserve proper names, numbers and terminology. Output only numbered lines in the format "1. translation".`
@@ -78,7 +106,7 @@ async function translateWithApi(items, target, config) {
78
106
  for (let start = 0; start < items.length; start += batchSize) {
79
107
  const batch = items.slice(start, start + batchSize)
80
108
  const content = batch.map((line, index) => `${index + 1}. ${line.text}`).join('\n')
81
- const response = await fetch(`${config.apiBase.replace(/\/+$/, '')}/chat/completions`, {
109
+ const response = await fetchWithRetry(`${config.apiBase.replace(/\/+$/, '')}/chat/completions`, {
82
110
  method: 'POST',
83
111
  headers: {
84
112
  'content-type': 'application/json',