@goodandready/dsh-voice 0.9.28 → 0.9.30

package/README.md

@@ -35,6 +35,9 @@
 
 ---
 
+## Requirements
+### Browser Microphone & Secure Context\nModern browsers require a **Secure Context** (`https://` or `http://localhost` / `http://127.0.0.1`) for microphone capture (`navigator.mediaDevices.getUserMedia`). When accessing DSH over a local network (e.g., `http://192.168.1.x:3080` via `dsh-lanmode`), browsers will block microphone access by default. Serve DSH behind an HTTPS reverse proxy or configure browser security flags (`chrome://flags/#unsafely-treat-insecure-origin-as-secure`) for LAN IP access.\n
+
 ## ⚡ Overview
 
 **`dsh-voice`** brings voice superpowers to the **DeepSeek Harness** Web UI. Whether you need hands-free real-time streaming dictation segmented on natural breath pauses or crisp voice notes with keyboard/mouse Push-to-Talk gestures, `dsh-voice` ensures your audio is never lost thanks to **automatic multi-provider fallback chains**.

package/README.ru.md

@@ -35,6 +35,9 @@
 
 ---
 
+## Требования
+### Доступ к микрофону и Secure Context\nСовременные веб-браузеры требуют безопасный контекст (**Secure Context**: `https://` или `http://localhost` / `http://127.0.0.1`) для захвата аудио с микрофона (`navigator.mediaDevices.getUserMedia`). При открытии DSH в локальной сети по обычному HTTP (например, `http://192.168.1.x:3080` через `dsh-lanmode`) браузер блокирует микрофон. Для работы в LAN настройте HTTPS (через reverse proxy, например caddy/nginx) либо укажите адрес в настройках браузера (`chrome://flags/#unsafely-treat-insecure-origin-as-secure`).\n
+
 ## ⚡ Обзор
 
 **`dsh-voice`** добавляет полноценные голосовые возможности в веб-интерфейс **DeepSeek Harness**. Будь то непрерывная диктовка с нарезкой фраз по естественным паузам или голосовые сообщения с удобными жестами Push-to-Talk (мышь и клавиатура) — `dsh-voice` гарантирует сохранность каждой записи благодаря **автоматическим цепочкам отказоустойчивости**.

package/README.zh.md

@@ -35,6 +35,9 @@
 
 ---
 
+## 运行要求
+### 浏览器麦克风与安全上下文 (Secure Context)\n现代浏览器要求在**安全上下文**（`https://` 或 `http://localhost` / `http://127.0.0.1`）下才允许调用麦克风接口（`navigator.mediaDevices.getUserMedia`）。当通过局域网以普通 HTTP 访问 DSH（如使用 `dsh-lanmode` 访问 `http://192.168.1.x:3080`）时，浏览器默认会阻止录音。在局域网环境下，请配置 HTTPS 反向代理，或在浏览器中启用标志（`chrome://flags/#unsafely-treat-insecure-origin-as-secure`）。\n
+
 ## ⚡ 插件概览
 
 **`dsh-voice`** 为 **DeepSeek Harness** Web 界面带来极速语音交互体验。无论是按自然停顿切分的流式听写，还是带撤回保护的语音消息以及键盘/鼠标 Push-to-Talk 对讲，`dsh-voice` 凭借**多服务商自动故障转移备用链**确保您的录音万无一失。

package/lib/client.js

@@ -1080,39 +1080,73 @@ window.__ModuleLoader__.load({
     // Fetch the mode chain from the host once. Only needed to decide whether
 
     voice._core = { tidyPhrase, applyVoiceCommands, VOICE_COMMANDS, undoLastInsert, insertHistory, insertDraftText, appendDraft, resolveBrowserRecognitionLang }
-// lib/client-src/35-webgpu.js — Client-side WebGPU Whisper offline recognition
-// Detects WebGPU availability in browser and provides offline transcription with graceful fallback.
+// lib/client-src/35-webgpu.js — Client-side WebGPU/WASM Whisper offline recognition
+// Detects WebGPU and WebAssembly availability in browser and provides offline transcription with graceful WASM fallback.
 
 function isWebGpuSupported() {
   if (typeof navigator === 'undefined' || !navigator.gpu) return false
   return true
 }
 
+function isWasmSupported() {
+  return typeof WebAssembly !== 'undefined'
+}
+
+function isOfflineSupported() {
+  return isWebGpuSupported() || isWasmSupported()
+}
+
 const _webGpuPipelines = new Map()
 
-function loadWebGpuWhisper(modelName) {
+function loadWebGpuWhisper(modelName, options) {
   const targetModel = modelName || (voice.settings && voice.settings.webgpuModel) || 'onnx-community/whisper-tiny'
-  if (_webGpuPipelines.has(targetModel)) {
-    return _webGpuPipelines.get(targetModel)
+  const forceDevice = options?.device
+  const key = `${targetModel}:${forceDevice || 'auto'}`
+  if (_webGpuPipelines.has(key)) {
+    return _webGpuPipelines.get(key)
   }
   const promise = (async () => {
-    if (!isWebGpuSupported()) {
-      throw new Error('WebGPU is not supported in this browser environment')
+    if (!isOfflineSupported()) {
+      throw new Error('Neither WebGPU nor WebAssembly is supported in this browser environment')
     }
     const mod = typeof globalThis !== 'undefined' && globalThis.__mockTransformers
       ? globalThis.__mockTransformers
       : (await import('https://cdn.jsdelivr.net/npm/@huggingface/transformers@3.3.3'))
     const pipeline = mod.pipeline || (mod.default && mod.default.pipeline)
     if (!pipeline) throw new Error('Transformers pipeline not found in loaded bundle')
-    return pipeline('automatic-speech-recognition', targetModel, {
-      device: 'webgpu',
-      dtype: 'fp32',
-    })
+
+    // Prefer WebGPU if available and not forced to wasm; fall back to WASM on error or when WebGPU is absent
+    const canGpu = forceDevice === 'webgpu' || (!forceDevice && isWebGpuSupported())
+    if (canGpu) {
+      try {
+        return await pipeline('automatic-speech-recognition', targetModel, {
+          device: 'webgpu',
+          dtype: 'fp32',
+        })
+      } catch (gpuErr) {
+        if (isWasmSupported()) {
+          return await pipeline('automatic-speech-recognition', targetModel, {
+            device: 'wasm',
+            dtype: 'q8',
+          })
+        }
+        throw gpuErr
+      }
+    }
+
+    if (isWasmSupported()) {
+      return await pipeline('automatic-speech-recognition', targetModel, {
+        device: 'wasm',
+        dtype: 'q8',
+      })
+    }
+
+    throw new Error('Offline speech recognition failed: no WebGPU or WebAssembly available')
   })().catch((err) => {
-    _webGpuPipelines.delete(targetModel)
+    _webGpuPipelines.delete(key)
     throw err
   })
-  _webGpuPipelines.set(targetModel, promise)
+  _webGpuPipelines.set(key, promise)
   return promise
 }
 
@@ -1131,10 +1165,10 @@ async function audioBlobToFloat32Array(blob) {
 }
 
 async function transcribeWithWebGpu(blob, mime, options) {
-  if (!isWebGpuSupported()) {
-    throw new Error('WebGPU unavailable in browser')
+  if (!isOfflineSupported()) {
+    throw new Error('Offline recognition (WebGPU/WASM) unavailable in browser')
   }
-  const transcriber = await loadWebGpuWhisper(options && options.model)
+  const transcriber = await loadWebGpuWhisper(options && options.model, options)
   const audioData = await audioBlobToFloat32Array(blob)
   const generateArgs = {}
   if (options && options.language && options.language !== 'auto') {
@@ -1146,7 +1180,9 @@ async function transcribeWithWebGpu(blob, mime, options) {
 }
 
 voice.webGpu = {
-  isSupported: isWebGpuSupported,
+  isSupported: isOfflineSupported,
+  isWebGpuSupported,
+  isWasmSupported,
   transcribe: transcribeWithWebGpu,
   load: loadWebGpuWhisper,
   clearCache: () => { _webGpuPipelines.clear() },
@@ -3051,11 +3087,13 @@ voice._chainsPromise = null
       const [updaterStatus, setUpdaterStatus] = React.useState(null)
       const [updating, setUpdating] = React.useState(false)
       const [updaterMsg, setUpdaterMsg] = React.useState('')
+      const [updaterMsgKind, setUpdaterMsgKind] = React.useState('')
       const [updaterLoading, setUpdaterLoading] = React.useState(false)
 
       const checkUpdater = React.useCallback(async () => {
         setUpdaterLoading(true)
         setUpdaterMsg('')
+        setUpdaterMsgKind('')
         try {
           const res = await fetch('/api/dsh-voice/update', {
             headers: { accept: 'application/json' },
@@ -3066,10 +3104,12 @@ voice._chainsPromise = null
           setUpdaterStatus(data)
           if (data && data.latestCheckFailed) {
             setUpdaterMsg(t('updateCheckFailed'))
+            setUpdaterMsgKind('err')
           }
         } catch (e) {
           setUpdaterStatus(null)
           setUpdaterMsg(t('updateCheckFailed') + ': ' + (e?.message || String(e)))
+          setUpdaterMsgKind('err')
         } finally {
           setUpdaterLoading(false)
         }
@@ -3079,6 +3119,7 @@ voice._chainsPromise = null
         if (updating) return
         setUpdating(true)
         setUpdaterMsg('')
+        setUpdaterMsgKind('')
         try {
           const res = await fetch('/api/dsh-voice/update', {
             method: 'POST',
@@ -3104,15 +3145,17 @@ voice._chainsPromise = null
             throw new Error(data.error || `HTTP ${res.status}`)
           }
           setUpdaterMsg(data.message || t('updateSuccess'))
+          setUpdaterMsgKind('ok')
           checkUpdater()
         } catch (err) {
           setUpdaterMsg(err && err.message ? err.message : String(err))
+          setUpdaterMsgKind('err')
         } finally {
           setUpdating(false)
         }
       }
 
-      return { updaterStatus, updating, updaterMsg, updaterLoading, checkUpdater, onUpdateNow }
+      return { updaterStatus, updating, updaterMsg, updaterMsgKind, updaterLoading, checkUpdater, onUpdateNow }
     }
     // Developer Lexicon & IT Jargon
     function JargonEditor(props) {
@@ -3249,6 +3292,7 @@ voice._chainsPromise = null
       const updaterLoading = props.updaterLoading
       const updating = props.updating
       const updaterMsg = props.updaterMsg
+      const updaterMsgKind = props.updaterMsgKind
       const onCheck = props.onCheck
       const onUpdateNow = props.onUpdateNow
 
@@ -3292,7 +3336,7 @@ voice._chainsPromise = null
               style: {
                 marginTop: '8px',
                 fontSize: '12px',
-                color: updaterMsg.includes('fail') || updaterMsg.includes('失败')
+                color: (updaterMsgKind === 'err' || (!updaterMsgKind && (updaterMsg.includes('fail') || updaterMsg.includes('失败'))))
                   ? 'var(--dsw-alias-state-warning-primary)'
                   : 'var(--dsw-alias-state-success-primary)',
               },
@@ -3495,7 +3539,7 @@ voice._chainsPromise = null
 
       const { testingMic, testLevel, toggleTestMic, gateDbVal, gateThresholdPercent } = useMicTester(draft, value, setErr)
 
-      const { updaterStatus, updating, updaterMsg, updaterLoading, checkUpdater, onUpdateNow } = usePluginUpdater()
+      const { updaterStatus, updating, updaterMsg, updaterMsgKind, updaterLoading, checkUpdater, onUpdateNow } = usePluginUpdater()
 
       const save = async () => {
         setErr(''); setSaved(false)
@@ -3869,6 +3913,7 @@ voice._chainsPromise = null
           updaterLoading: updaterLoading,
           updating: updating,
           updaterMsg: updaterMsg,
+          updaterMsgKind: updaterMsgKind,
           onCheck: checkUpdater,
           onUpdateNow: onUpdateNow,
         }),
@@ -3997,6 +4042,9 @@ voice._chainsPromise = null
       stopCurrent,
       submitPending,
       loadWebGpuWhisper,
+      isWebGpuSupported,
+      isWasmSupported,
+      isOfflineSupported,
       waitStop,
       currentLevel,
       accentColor,

package/lib/local-daemon.js

@@ -37,7 +37,7 @@ export function createLocalDaemons({ live, ctx }) {
       }
       const command = `${JSON.stringify(cfg.whisperBin)} -m ${JSON.stringify(cfg.whisperModel)}`
         + ` --host 127.0.0.1 --port 8001 -t 8 -p 1 -l ${isAutoLang(cfg.dictation.language) ? 'auto' : cfg.dictation.language}`
-      const specOpts = { command, onExpiry: 'none', timeoutMs: 0, stdoutMaxBytes: 4 * 1024 * 1024 }
+      const specOpts = { command, onExpiry: 'none', timeoutMs: 60000, stdoutMaxBytes: 4 * 1024 * 1024 }
       const spec = typeof ctx.shell?.resolve === 'function' ? ctx.shell.resolve(specOpts) : specOpts
 
       if (typeof ctx.shell?.execute === 'function') {
@@ -216,7 +216,7 @@ export function createLocalDaemons({ live, ctx }) {
       try { port = new URL(cfg.sensevoiceUrl).port || '6006' } catch { /* invalid URL */ }
       const cmdArgs = buildSensevoiceArgs(cfg, port)
       const commandStr = `${JSON.stringify(cfg.sensevoiceBin)}${cmdArgs}`
-      const specOpts = { command: commandStr, onExpiry: 'none', timeoutMs: 0, stdoutMaxBytes: 4 * 1024 * 1024 }
+      const specOpts = { command: commandStr, onExpiry: 'none', timeoutMs: 60000, stdoutMaxBytes: 4 * 1024 * 1024 }
       const spec = typeof ctx.shell?.resolve === 'function' ? ctx.shell.resolve(specOpts) : specOpts
 
       if (typeof ctx.shell?.execute === 'function') {

package/lib/sensevoice-installer.js

@@ -249,8 +249,13 @@ export function resolveSensevoiceProvider(cfg) {
       return 'cpu'
     }
     if (ext === '.rknn') {
-      return 'rknn'
+      if (modelPath && existsSync(modelPath)) {
+        return 'rknn'
+      }
+      // Guard: missing .rknn model file cannot be claimed as running accelerator (#168)
+      return 'cpu'
     }
+    return 'cpu'
   }
 
   return normProv || 'cpu'
@@ -263,10 +268,9 @@ export function buildSensevoiceArgs(cfg, port) {
     ? cfg.sensevoiceTokens
     : findTokensFile(modelDir)
   const tokensArg = tokensPath ? ` --tokens=${JSON.stringify(tokensPath)}` : ''
-  const rawProv = String(cfg?.sensevoiceProvider || '').trim().toLowerCase()
-  let normProv = rawProv === 'rknpu' ? 'rknn' : rawProv
+  const effectiveProv = resolveSensevoiceProvider(cfg)
 
-  if (normProv === 'rknn') {
+  if (effectiveProv === 'rknn') {
     const ext = path.extname(modelPath).toLowerCase()
     if (ext === '.onnx') {
       const rknnCandidate = path.join(modelDir, path.basename(modelPath, ext) + '.rknn')
@@ -275,15 +279,12 @@ export function buildSensevoiceArgs(cfg, port) {
         modelPath = rknnCandidate
       } else if (existsSync(defaultRknn)) {
         modelPath = defaultRknn
-      } else {
-        // Guard: ordinary ONNX model cannot be loaded by sherpa-onnx RKNN runtime (#221)
-        normProv = 'cpu'
       }
     }
   }
 
-  const provider = (normProv && normProv !== 'cpu')
-    ? ` --provider=${normProv}`
+  const provider = (effectiveProv && effectiveProv !== 'cpu')
+    ? ` --provider=${effectiveProv}`
     : ''
   const threads = Number(cfg?.sensevoiceThreads) || 4
   return ` --sense-voice-model=${JSON.stringify(modelPath)}`

package/lib/updater.js

@@ -18,6 +18,17 @@ function header(request, name) {
   return Array.isArray(value) ? value[0] : value;
 }
 
+function isPrivateLan(value) {
+  const address = value?.toLowerCase().replace(/^\[|\]$/g, '');
+  if (!address) return false;
+  const ipv4 = address.startsWith('::ffff:') ? address.slice(7) : address;
+  return (
+    ipv4.startsWith('192.168.') ||
+    ipv4.startsWith('10.') ||
+    /^172\.(1[6-9]|2[0-9]|3[0-1])\./.test(ipv4)
+  );
+}
+
 function isLoopback(value) {
   const address = value?.toLowerCase().replace(/^\[|\]$/g, '');
   return address === 'localhost' || address === 'localhost.' || address === '::1'
@@ -27,7 +38,8 @@ function isLoopback(value) {
 
 export function isTrustedUpdateRequest(request) {
   if (header(request, UPDATE_HEADER) !== '1') return false;
-  if (!isLoopback(request.socket?.remoteAddress)) return false;
+  const remote = request.socket?.remoteAddress;
+  if (!isLoopback(remote) && !isPrivateLan(remote)) return false;
   const site = header(request, 'sec-fetch-site');
   if (site !== undefined && site !== 'same-origin') return false;
   const origin = header(request, 'origin');
@@ -35,8 +47,11 @@ export function isTrustedUpdateRequest(request) {
   if (origin === undefined || host === undefined) return false;
   try {
     const url = new URL(origin);
-    return (url.protocol === 'http:' || url.protocol === 'https:')
-      && isLoopback(url.hostname) && url.host === host;
+    return (
+      (url.protocol === 'http:' || url.protocol === 'https:') &&
+      (isLoopback(url.hostname) || isPrivateLan(url.hostname)) &&
+      url.host === host
+    );
   } catch {
     return false;
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-voice",
-  "version": "0.9.28",
+  "version": "0.9.30",
   "description": "Voice input for DeepSeek Harness: dictation chunked by pauses and voice messages, each with its own provider fallback chain (Deepgram, Groq, HuggingFace, local whisper.cpp, plus any OpenAI-compatible API of your own).",
   "license": "MIT",
   "type": "module",