@goodandready/dsh-voice 0.9.27 → 0.9.29

package/README.md

@@ -219,7 +219,7 @@ Registers `transcribe_audio(file_path, language?)` in `ctx.tools`, allowing agen
 ### Internal HTTP Endpoints
 * `POST /dsh-voice/transcribe` — `{ dataBase64, mimeType, mode }` → `{ ok, text, provider, tookMs }`
 * `POST /dsh-voice/polish` — `{ text }` → `{ ok, text }`
-* `GET /dsh-voice/status` — Returns daemon status, active chains, SenseVoice.
+* `GET /dsh-voice/status` — Returns daemon status, active chains, SenseVoice and `effectiveSensevoiceProvider`.
 * `GET /dsh-voice/config` — Returns live plugin configuration snapshot.
 * `PUT /dsh-voice/config` — Updates and persists plugin configuration across network.
 * `GET/POST /dsh-voice/sensevoice-installer` — SenseVoice 1-click model installation status and trigger (protected by `isTrustedCaller`, model paths redacted).

package/README.ru.md

@@ -219,7 +219,7 @@ dsh plugin --profile web add @goodandready/dsh-voice
 ### Внутренние HTTP эндпоинты
 * `POST /dsh-voice/transcribe` — `{ dataBase64, mimeType, mode }` → `{ ok, text, provider, tookMs }`
 * `POST /dsh-voice/polish` — `{ text }` → `{ ok, text }`
-* `GET /dsh-voice/status` — состояние демонов, цепочки, конфигурация SenseVoice и реалтайма.
+* `GET /dsh-voice/status` — состояние демонов, цепочки, конфигурация SenseVoice, реалтайма и effectiveSensevoiceProvider.
 * `GET/PUT /dsh-voice/config` — получение и сохранение конфигурации плагина.
 * `GET/POST /dsh-voice/sensevoice-installer` — статус и установка модели SenseVoice в 1 клик.
 * `POST /api/dsh-voice/update` — запуск обновления плагина до актуальной версии из npm (с проверкой прав и блокировки профиля).

package/README.zh.md

@@ -138,7 +138,7 @@ v0.8.18 质量审查后的修复批次（Gitea #79–#87，PR #88）：
 
 * `POST /dsh-voice/transcribe` — 音频转写：`{ dataBase64, mimeType, mode }` → `{ ok, text, provider, tookMs }`
 * `POST /dsh-voice/polish` — 文本模型润色：`{ text }` → `{ ok, text }`
-* `GET /dsh-voice/status` — 查询后端状态、服务商健康度与 SenseVoice
+* `GET /dsh-voice/status` — 查询后端状态、服务商健康度、SenseVoice 及 effectiveSensevoiceProvider
 * `GET /dsh-voice/config` — 获取插件运行时配置快照
 * `PUT /dsh-voice/config` — 跨网络安全持久化更新插件配置
 * `GET/POST /dsh-voice/sensevoice-installer` — SenseVoice 一键模型安装状态与触发接口（受 `isTrustedCaller` 保护，绝对路径脱敏）

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() },
@@ -3997,6 +4033,13 @@ voice._chainsPromise = null
       stopCurrent,
       submitPending,
       loadWebGpuWhisper,
+      isWebGpuSupported,
+      isWasmSupported,
+      isOfflineSupported,
+      waitStop,
+      currentLevel,
+      accentColor,
+      softColor,
     }
     return module.exports
   },

package/lib/host-routes.js

@@ -1,5 +1,6 @@
 // lib/host-routes.js — Host HTTP routes for dsh-voice status, polish, and transcribe
 import { writeJson, readBody, isTrustedCaller } from './http-util.js'
+import { resolveSensevoiceProvider } from './sensevoice-installer.js'
 
 export function registerHostRoutes(ctx, {
   live,
@@ -63,6 +64,7 @@ export function registerHostRoutes(ctx, {
         sensevoiceUrl: cfg.sensevoiceUrl,
         sensevoiceAutostart: !!cfg.sensevoiceAutostart,
         sensevoiceProvider: cfg.sensevoiceProvider || 'cpu',
+        effectiveSensevoiceProvider: resolveSensevoiceProvider(cfg),
         sensevoiceThreads: cfg.sensevoiceThreads || 4,
         webgpuModel: cfg.webgpuModel || 'onnx-community/whisper-tiny',
         deepgramBaseUrl: cfg.deepgramBaseUrl || 'https://api.deepgram.com',

package/lib/index.js

@@ -47,6 +47,23 @@ export const Config = typeof z.number().volatile === 'function'
   ? makeVolatile(BaseConfig)
   : BaseConfig
 
+export function isVolatileRef(value) {
+  return value !== null && typeof value === 'object' && !Array.isArray(value) && typeof value.get === 'function'
+}
+
+export function plainConfig(value) {
+  while (typeof value === 'function') {
+    value = value()
+  }
+  if (isVolatileRef(value)) return plainConfig(value.get())
+  if (value === null || typeof value !== 'object') return value
+  if (value instanceof Date || value instanceof RegExp || (typeof Buffer !== 'undefined' && Buffer.isBuffer(value))) {
+    return value
+  }
+  if (Array.isArray(value)) return value.map(plainConfig)
+  return Object.fromEntries(Object.entries(value).map(([k, v]) => [k, plainConfig(v)]))
+}
+
 export function apply(ctx, baseConfig) {
   let settingsService = null
   let scope = null
@@ -55,7 +72,25 @@ export function apply(ctx, baseConfig) {
   const readEntryConfig = createConfigReader(baseConfig, ctx)
   getConfig = readEntryConfig
 
-  const live = () => BaseConfig(structuredClone(getConfig() ?? {})) ?? readEntryConfig()
+  const describeRow = () => {
+    if (typeof settingsService?.describe !== 'function') return null
+    try {
+      return settingsService.describe().find((r) => r && r.ns === NS) || null
+    } catch {
+      return null
+    }
+  }
+
+  const live = () => {
+    try {
+      const raw = getConfig() ?? {}
+      const unwrapped = plainConfig(raw)
+      const validated = BaseConfig(unwrapped)
+      return plainConfig(validated) ?? plainConfig(readEntryConfig())
+    } catch {
+      return plainConfig(readEntryConfig())
+    }
+  }
 
   const resolveKey = async (ref) => {
     if (!ref || !ref.type) return ''
@@ -81,14 +116,19 @@ export function apply(ctx, baseConfig) {
     try {
       scope = sctx.settings.register(NS, Config, { base: baseConfig })
       if (scope && typeof scope.get === 'function') {
-        getConfig = () => scope.get() ?? readEntryConfig()
+        getConfig = () => scope.get() ?? describeRow()?.config ?? readEntryConfig()
+      } else {
+        getConfig = () => describeRow()?.config ?? readEntryConfig()
       }
       if (scope && typeof scope.watch === 'function') {
         scope.watch(() => {
           triggerAutostart()
         })
       }
-    } catch { /* modern DSH handles settings via SettingsForms */ }
+    } catch {
+      // Modern DSH 0.2 handles settings via SettingsForms and describe/update/replace
+      getConfig = () => describeRow()?.config ?? readEntryConfig()
+    }
     sctx.effect(() => () => { getConfig = readEntryConfig; scope = null; settingsService = null })
     triggerAutostart()
   })

package/lib/sensevoice-installer.js

@@ -231,6 +231,36 @@ export function registerSensevoiceInstaller(ctx, { liveConfig, isAlive, updateCo
   })
 }
 
+export function resolveSensevoiceProvider(cfg) {
+  let modelPath = String(cfg?.sensevoiceModel || '')
+  const rawProv = String(cfg?.sensevoiceProvider || '').trim().toLowerCase()
+  let normProv = rawProv === 'rknpu' ? 'rknn' : rawProv
+
+  if (normProv === 'rknn') {
+    const ext = path.extname(modelPath).toLowerCase()
+    if (ext === '.onnx') {
+      const modelDir = modelPath ? path.dirname(modelPath) : getDefaultModelDir()
+      const rknnCandidate = path.join(modelDir, path.basename(modelPath, ext) + '.rknn')
+      const defaultRknn = path.join(modelDir, 'model.rknn')
+      if (existsSync(rknnCandidate) || existsSync(defaultRknn)) {
+        return 'rknn'
+      }
+      // Guard: ordinary ONNX model cannot be loaded by sherpa-onnx RKNN runtime (#221)
+      return 'cpu'
+    }
+    if (ext === '.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'
+}
+
 export function buildSensevoiceArgs(cfg, port) {
   let modelPath = String(cfg?.sensevoiceModel || '')
   const modelDir = modelPath ? path.dirname(modelPath) : getDefaultModelDir()
@@ -238,10 +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')
@@ -250,15 +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/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-voice",
-  "version": "0.9.27",
+  "version": "0.9.29",
   "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",