dsh-voice-mode 0.1.0

package/package.json

@@ -0,0 +1,89 @@
+{
+  "name": "dsh-voice-mode",
+  "description": "Full-duplex voice conversation mode for DeepSeek Harness: streamed zipformer2 ASR to draft, Edge TTS sentence-by-sentence reading of final replies, voice barge-in · DSH 语音双工对话模式",
+  "version": "0.1.0",
+  "type": "module",
+  "main": "lib/index.js",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/qishuilalala/dsh-voice-mode.git"
+  },
+  "homepage": "https://github.com/qishuilalala/dsh-voice-mode#readme",
+  "bugs": {
+    "url": "https://github.com/qishuilalala/dsh-voice-mode/issues"
+  },
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "exports": {
+    ".": "./lib/index.js",
+    "./client": "./lib/client.js",
+    "./cordis.patch.yml": "./cordis.patch.yml",
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "lib/index.js",
+    "lib/client.js",
+    "cordis.patch.yml",
+    "assets/architecture.svg",
+    "assets/demo.gif",
+    "scripts/prefetch.mjs",
+    "scripts/list-voices.mjs",
+    "README.md",
+    "README.en.md",
+    "LICENSE"
+  ],
+  "scripts": {
+    "build": "node build.mjs",
+    "prepack": "node build.mjs",
+    "test": "node test/segmenter.test.mjs && node test/wakeword.test.mjs && node test/verify-client.mjs",
+    "verify:client": "node test/verify-client.mjs",
+    "prefetch": "node scripts/prefetch.mjs",
+    "list-voices": "node scripts/list-voices.mjs",
+    "verify": "node scripts/verify.mjs",
+    "typecheck": "node node_modules/typescript/bin/tsc -p tsconfig.json --noEmit && node node_modules/typescript/bin/tsc -p tsconfig.client.json --noEmit"
+  },
+  "dsh": {
+    "bundle": {
+      "patch": "./cordis.patch.yml"
+    },
+    "client": {
+      "platform": "web",
+      "inject": [
+        "@deepseek-ai/dsh-client-runtime",
+        "@deepseek-ai/dsh-client-ui-slots"
+      ]
+    }
+  },
+  "dependencies": {
+    "msedge-tts": "^2.0.7",
+    "sherpa-onnx": "^1.13.6",
+    "@deepseek-ai/schemastery": "^3.18.1"
+  },
+  "peerDependencies": {
+    "@deepseek-ai/cordis": "^4.0.1",
+    "@deepseek-ai/dsh-web": "^0.1.0-rc.6 || ^0.1.1-rc.0",
+    "react": "^18.2.0"
+  },
+  "devDependencies": {
+    "esbuild": "^0.25.0",
+    "typescript": "^5.9.2",
+    "@types/node": "^26.0.0",
+    "@types/react": "^19.0.0"
+  },
+  "keywords": [
+    "dsh",
+    "dsh-plugin",
+    "voice",
+    "speech",
+    "asr",
+    "tts",
+    "duplex",
+    "zipformer2",
+    "msedge-tts"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "license": "MIT"
+}

package/scripts/list-voices.mjs

@@ -0,0 +1,29 @@
+#!/usr/bin/env node
+/**
+ * 打印 msedge-tts 可用的音色清单（README 音色参考表的权威来源）。
+ *
+ * 用法：
+ *   node scripts/list-voices.mjs            # 全部音色
+ *   node scripts/list-voices.mjs zh         # 仅中文（zh-CN / zh-HK / zh-TW）
+ *   node scripts/list-voices.mjs zh,en      # 多区域过滤（逗号分隔前缀）
+ */
+import { MsEdgeTTS } from 'msedge-tts'
+
+const filter = (process.argv[2] ?? '').split(',').filter(Boolean).map((s) => s.trim().toLowerCase())
+
+const tts = new MsEdgeTTS()
+const voices = await tts.getVoices()
+const rows = voices
+  .filter((v) => filter.length === 0 || filter.some((f) => v.Locale.toLowerCase().startsWith(f)))
+  .sort((a, b) => a.Locale.localeCompare(b.Locale) || a.ShortName.localeCompare(b.ShortName))
+  .map((v) => `${v.ShortName}\t${v.Locale}\t${v.Gender}\t${v.FriendlyName}`)
+
+if (rows.length === 0) {
+  console.error('未获取到音色（网络不可达或服务变更），请检查后重试。')
+  process.exitCode = 1
+} else {
+  console.log(`共 ${rows.length} 个音色（ShortName 即配置 voice 的取值）：`)
+  console.log('ShortName\tLocale\tGender\tFriendlyName')
+  console.log(rows.join('\n'))
+}
+await tts.close()

package/scripts/prefetch.mjs

@@ -0,0 +1,97 @@
+#!/usr/bin/env node
+/**
+ * dsh-voice-mode 模型预下载（prefetch）：安装后、首次使用前预热 zipformer2
+ * 模型缓存，避免第一次进入语音模式等待下载（约 160MB）。
+ *
+ * 与运行时代码保持一致的约定：
+ *  - 缓存目录平台默认值：Windows = %LOCALAPPDATA%\dsh-voice-mode\models；
+ *    类 Unix = ~/.cache/dsh-voice-mode/models（src/index.ts defaultModelCacheDir）。
+ *  - 模型清单与镜像回退（huggingface.co -> hf-mirror.com）与 src/asr-host.ts
+ *    的 MODEL_REPO / MODEL_FILES / HOST_PRIMARY / HOST_FALLBACK 保持一致。
+ *
+ * 用法：node scripts/prefetch.mjs [--cache-dir <path>]
+ */
+import { createWriteStream } from 'node:fs'
+import { mkdir, rename, stat, unlink } from 'node:fs/promises'
+import { homedir } from 'node:os'
+import { join } from 'node:path'
+
+// ---- 与 src/asr-host.ts 同步的模型常量（改动时需两处一致）----
+const MODEL_REPO = 'csukuangfj/sherpa-onnx-streaming-zipformer-zh-int8-2025-06-30'
+const MODEL_FILES = ['encoder.int8.onnx', 'decoder.onnx', 'joiner.int8.onnx', 'tokens.txt']
+const HOST_PRIMARY = 'https://huggingface.co'
+const HOST_FALLBACK = 'https://hf-mirror.com'
+
+const defaultCacheDir = () =>
+  process.platform === 'win32'
+    ? join(process.env.LOCALAPPDATA ?? join(homedir(), 'AppData', 'Local'), 'dsh-voice-mode', 'models')
+    : join(homedir(), '.cache', 'dsh-voice-mode', 'models')
+
+const argCache = process.argv.indexOf('--cache-dir')
+const cacheDir = argCache !== -1 ? process.argv[argCache + 1] ?? defaultCacheDir() : defaultCacheDir()
+const repoDir = join(cacheDir, MODEL_REPO)
+
+async function downloadFile(repoDir, file) {
+  const localPath = join(repoDir, file)
+  if ((await stat(localPath).catch(() => null))?.isFile()) {
+    console.log(`  ✓ ${file} 已存在`)
+    return true
+  }
+  const partPath = `${localPath}.part`
+  const partSt = await stat(partPath).catch(() => null)
+  for (const host of [HOST_PRIMARY, HOST_FALLBACK]) {
+    const url = `${host}/${MODEL_REPO}/resolve/main/${file}`
+    const headers = { 'user-agent': 'dsh-voice-mode/prefetch' }
+    const resumeFrom = partSt?.size ?? 0
+    if (resumeFrom > 0) headers.range = `bytes=${resumeFrom}-`
+    try {
+      const res = await fetch(url, { headers })
+      if (res.status === 416) {
+        await rename(partPath, localPath).catch(() => undefined)
+        console.log(`  ✓ ${file}（续传完成）`)
+        return true
+      }
+      if (res.status !== 200 && res.status !== 206) continue
+      const total = Number(res.headers.get('content-length') ?? 0) + resumeFrom
+      const statusMax = file.length + 18
+      const sink = createWriteStream(partPath, resumeFrom > 0 ? { flags: 'a' } : {})
+      const reader = res.body
+      if (!reader) continue
+      let received = resumeFrom
+      let lastPct = -1
+      for (;;) {
+        const { done, value } = await reader.read()
+        if (done) break
+        received += value.byteLength
+        if (!sink.write(value)) await new Promise((r) => sink.once('drain', r))
+        const pct = total > 0 ? Math.min(100, Math.round((received / total) * 100)) : -1
+        if (pct >= 0 && pct !== lastPct && pct % 10 === 0) {
+          lastPct = pct
+          process.stdout.write(`\r  ${file.padEnd(statusMax)} ${pct}%`)
+        }
+      }
+      await new Promise((resolve, reject) => {
+        sink.end(resolve)
+        sink.on('error', reject)
+      })
+      await rename(partPath, localPath)
+      process.stdout.write(`\r  ${file.padEnd(statusMax)} 100%\n`)
+      console.log(`  ✓ ${file} 完成（${(received / 1048576).toFixed(1)} MB）`)
+      return true
+    } catch {
+      // 换下一个 host
+    }
+  }
+  await unlink(partPath).catch(() => undefined)
+  console.error(`  ✗ ${file} 下载失败（两个镜像均不可达）`)
+  return false
+}
+
+console.log(`模型缓存目录：${repoDir}`)
+await mkdir(repoDir, { recursive: true })
+let ok = true
+for (const f of MODEL_FILES) {
+  if (!(await downloadFile(repoDir, f))) ok = false
+}
+console.log(ok ? '\n预下载完成，可直接进入语音模式。' : '\n预下载未完成，可稍后重试（断点续传）。')
+process.exitCode = ok ? 0 : 1