@goodandready/dsh-voice 0.9.44 → 0.9.45

package/lib/client.js

@@ -1,13 +1,6 @@
 // dsh-voice — client half (browser).
 //
-// Two buttons in conversation.input.right:
-//   mic  — dictation: speech is cut on pauses, each chunk is recognized and
-//          appended to the composer; send stays with the user;
-//   wave — voice message: one whole recording; after recognition the text is
-//          sent to the agent when the cancel window expires.
-//
-// The recording pill lives in conversation.input.dock; settings are a card
-// in settings.plugin.item.
+// Dictation (mic), voice message (wave), dock pill and settings card.
 
 window.__ModuleLoader__.load({
   id: '@goodandready/dsh-voice',
@@ -32,9 +25,7 @@ window.__ModuleLoader__.load({
     const ROW_ID = 'dsh-voice'
     const ROW_CONFIG_KEY = PKG + '#' + ROW_ID
 
-    // UI strings live in the locale registry so a separate package can
-    // translate them without touching this plugin. English is the source
-    // language, the default, and the fallback.
+    // UI strings in locale registry; English is source/default.
     const en = {
 'saveFailed': 'Some fields were not saved —',
 'cardHint': 'Dictation and voice messages: providers, chains, local whisper',
@@ -320,6 +311,7 @@ window.__ModuleLoader__.load({
 'wizardTestFailed': 'STT test failed: {error}',
 'wizardNextStepHint': 'Next step: {hint}',
 'wizardHintPermission': 'Click "Request Microphone Access" and allow browser permission prompt.', 'wizardHintDevice': 'Check microphone connection or select a different input device.', 'wizardHintHttps': 'Open DeepSeek Harness via HTTPS or localhost address.', 'wizardHintProvider': 'Configure STT engine credentials or install SenseVoice model below.',
+'wizardHintWinDesktop': 'System microphone blocked. Open Windows Settings → Privacy & security → Microphone, and turn ON "Let desktop apps access your microphone".', 'wizardHintWinBrowser': 'Microphone permission denied. Allow access in your browser address bar or Windows Settings → Privacy & security → Microphone.', 'wizardHintMacDesktop': 'System microphone blocked. Open macOS System Settings → Privacy & Security → Microphone, and enable DSH Desktop.', 'wizardOpenWinMicSettings': 'Open Windows Mic Settings',
 'wizardFinishSuccess': 'Setup & Verification Complete', 'wizardSkip': 'Skip Wizard',
 'wizardStatusCompleted': 'Microphone and STT verified',
 'exportDiagnosticsBtn': 'Export Diagnostics Report', 'exportingDiagnostics': 'Exporting report…',
@@ -612,6 +604,7 @@ window.__ModuleLoader__.load({
 'wizardTestFailed': 'STT 测试失败：{error}',
 'wizardNextStepHint': '建议下一步：{hint}',
 'wizardHintPermission': '点击“请求麦克风权限”并在浏览器弹窗中允许访问。', 'wizardHintDevice': '检查麦克风硬件连接或切换选择其他输入设备。', 'wizardHintHttps': '请通过 HTTPS 或 localhost 访问 DeepSeek Harness。', 'wizardHintProvider': '请在下方配置 STT 密钥凭据或一键安装 SenseVoice 本地模型。',
+'wizardHintWinDesktop': '系统麦克风被阻止。请在 Windows【设置】→【隐私和安全性】→【麦克风】中开启最下方的【允许桌面应用访问你的麦克风】。', 'wizardHintWinBrowser': '麦克风权限被拒绝。请在浏览器地址栏允许麦克风权限，或在 Windows【设置】→【隐私和安全性】→【麦克风】中开启访问。', 'wizardHintMacDesktop': '系统麦克风被阻止。请在 macOS【系统设置】→【隐私与安全性】→【麦克风】中勾选允许 DSH Desktop。', 'wizardOpenWinMicSettings': '打开系统麦克风设置',
 'wizardFinishSuccess': '设置与验证已完成', 'wizardSkip': '跳过向导',
 'wizardStatusCompleted': '麦克风与 STT 诊断已就绪',
 'exportDiagnosticsBtn': '导出诊断报告', 'exportingDiagnostics': '正在导出报告…',
@@ -1184,8 +1177,7 @@ window.__ModuleLoader__.load({
         if (resolveStop) { resolveStop(); resolveStop = null }
         options.onError(code || t('recognitionError'))
       }
-      // The browser ends recognition on its own (pauses, timeout). Restart
-      // until we stop it, otherwise dictation dies silently on the first pause.
+      // Restart browser recognition until explicitly stopped
       recognition.onend = () => {
         if (stopped) {
           if (resolveStop) { resolveStop(); resolveStop = null }
@@ -1219,8 +1211,7 @@ window.__ModuleLoader__.load({
     // Fetch the mode chain from the host once. Only needed to decide whether
 
     voice._core = { tidyPhrase, applyVoiceCommands, VOICE_COMMANDS, undoLastInsert, undoLastChunk: undoLastInsert, insertHistory, insertDraftText, appendDraft, resolveBrowserRecognitionLang }
-// 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.
+// Client WebGPU/WASM Whisper offline recognition (#168)
 
 function isWebGpuSupported() {
   if (typeof navigator === 'undefined' || !navigator.gpu) return false
@@ -1326,8 +1317,7 @@ voice.webGpu = {
   load: loadWebGpuWhisper,
   clearCache: () => { _webGpuPipelines.clear() },
 }
-// lib/client-src/36-silero-vad.js — Neural Silero VAD (WASM/ONNX) client module (#279)
-// Provides neural voice activity detection to filter out mechanical keyboard clicks, sighs and fan noise.
+// Neural Silero VAD client module (#279)
 
 function isSileroVadSupported() {
   return typeof Float32Array !== 'undefined'
@@ -1373,8 +1363,7 @@ voice.sileroVad = {
   evaluate: evaluateSileroFrame,
   reset: () => { _sileroState = { h: null, c: null } },
 }
-// lib/client-src/37-ws-stream.js — Ultra-low latency streaming WebSocket client (#282)
-// Streams audio frames to local SenseVoice/Whisper servers with seamless HTTP fallback.
+// Streaming WebSocket client (#282)
 
 function isWsStreamingSupported() {
   return typeof WebSocket !== 'undefined'
@@ -2146,9 +2135,7 @@ voice._chainsPromise = null
       return String(name).replace(/^Key/, '').replace(/^Digit/, '')
     }
 
-    // What to store on key press. Pure modifiers are remembered by name:
-    // left and right have different codes but the user means "either".
-    // Supports combos like Control+Space or Alt+KeyV.
+    // Hotkey combo normalization
     function keyFromEvent(event) {
       if (!event) return ''
       const modifiers = []
@@ -2211,11 +2198,7 @@ voice._chainsPromise = null
       return event.code === name || event.key === name
     }
 
-    // Announce that the user started speaking.
-    //
-    // Playback should mute immediately: listening and talking at once is
-    // impossible. There is no plugin-to-plugin API — this broadcasts a window
-    // event that anyone may hear, so either side works alone.
+    // Broadcast window event so active playback mutes when speaking.
 
 // --------------------------------------------------------- hold gesture
     const HOLD_THRESHOLD_MS = 350
@@ -2332,8 +2315,7 @@ voice._chainsPromise = null
       }
       return cleanup
     }
-// ------------------------------------------------------------------ 43-preview
-// Editable preview state machine, session pinning & UI component (#170).
+// Editable preview state machine (#170)
 
 function initPendingPreview(text, target, options = {}) {
   const autoMs = Number(options.autoSendMs) || 4000
@@ -2787,16 +2769,13 @@ function cycleDictationMode() {
           const adapt = Number(voice.settings.vadAdapt) || 0
           let effectiveVad = Number(voice.settings.vadSilenceMs) || 700
           if (adapt > 0 && rec.hadSpeech) {
-            // Adaptive threshold (#41): speech density over ~1s (20 samples).
-            // Dense speech -> lower threshold (cut more precisely);
-            // pause-heavy -> threshold rises toward base (do not cut on breaths).
+            // Adaptive threshold (#41)
             const win = voice.levels.slice(-20)
             const density = win.length ? win.filter((v) => v > 0.06).length / win.length : 0
             const k = adapt * (density - 0.5) * 2
             effectiveVad = Math.max(150, Math.round(Number(voice.settings.vadSilenceMs) * (1 - k)))
           }
-          // Continuous mode (#40): cut on a timer while speech continues,
-          // without waiting for a long pause.
+          // Continuous mode (#40)
           const stream = !!voice.settings.stream && rec.mode === 'dictation'
           if (stream && rec.hadSpeech && !rec.cutting) {
             rec.streamMs += tick
@@ -2888,8 +2867,7 @@ function cycleDictationMode() {
           const p = voice.pending
           if (!p) return
           if (p.undoOnly) {
-            // Undo-only mode (#29-5): when the window expires just hide the
-            // panel; the text either stayed or was already undone.
+            // Undo-only mode (#29-5)
             p.leftMs -= tick
             if (p.leftMs <= 0) voice.set({ phase: 'idle' })
             else voice.notify()
@@ -2947,8 +2925,7 @@ function cycleDictationMode() {
         const inBrowser = !!voice.browser
         const rec = voice.rec
         const hint = v.mode === 'dictation' ? t('dictationPill') : t('messagePill')
-        // Live caption of what is heard right now. Until the browser emits a
-        // final chunk the text is interim and changes on screen.
+        // Live caption: interim preview
         const caption = voice.caption || (inBrowser ? '' : null)
         const activeSubMode = (voice._activeTarget && voice._activeTarget.dictationMode)
           || voice.selectedDictationMode
@@ -3112,8 +3089,7 @@ function cycleDictationMode() {
 
     // --------------------------------------------------------------- slots
     function registerComposer(ctx) {
-      // The hotkey lives for as long as the plugin is applied and can change
-      // without restart: the settings card broadcasts and the composer reloads.
+      // Hotkey listener with dynamic reload
       ctx.effect(() => {
         let dispose = () => {}
         let alive = true
@@ -3176,9 +3152,7 @@ function cycleDictationMode() {
         }
       }, 'dsh-voice: hold hotkey')
 
-      // locale: NS is required — the core resolves `label` through the inject face,
-      // and without it the label throws "cannot get property t without inject" while
-      // the slot snapshot is built, breaking the whole client batch.
+      // locale: NS required for core slot snapshot injection
       ctx.effect(() => {
         const offRight = ctx.slots.inject('conversation.input.right', () => ctx.slots.register(
           { name: 'conversation.input.right', id: '@goodandready/dsh-voice', order: 6, locale: NS, label: () => t('title') },
@@ -4458,11 +4432,7 @@ function MacrosSection(props) {
   function VoiceSection(props) {
    const t = (props && props.t) || moduleT
    const ctx = (props && props.ctx) || {}
-   // On a non-localhost page the kernel disables settings entirely: the
-   // shared document is not readable, every section gets "unavailable" and
-   // writes are dropped. The server does not share that restriction, and
-   // dsh-lanmode rebuilds the same mechanics on the same calls. Prefer its
-   // service when installed, otherwise the kernel one (fine on loopback).
+   // Settings adapter: prefer dsh-lanmode when remote, otherwise kernel.
    const settingsService = (ctx.get && ctx.get('lanSettings')) || ctx.configForms || ctx.settingsScope
    const scope = settingsService && typeof settingsService.get === 'function'
     ? settingsService.get(NS)
@@ -4778,10 +4748,7 @@ function MacrosSection(props) {
      onOpenWizard: () => setShowWizard((v) => !v),
     }),
     showWizard ? h(FirstRunWizard, {
-     t: t,
-     draft: draft,
-     setTop: setTop,
-     writable: writable,
+     t, draft, setTop, writable,
      onClose: () => setShowWizard(false),
     }) : null,
     h('div', { className: 'cb-section-card' },
@@ -4877,10 +4844,7 @@ function MacrosSection(props) {
      }),
     ),
     h(HardwareOptionsCard, {
-     t: t,
-     draft: draft,
-     setTop: setTop,
-     writable: writable,
+     t, draft, setTop, writable,
      catching: catching,
      setCatching: setCatching,
      devices: devices,
@@ -4976,23 +4940,14 @@ function MacrosSection(props) {
     ),
     // Dictation Modes & Prompts (#264)
     h(DictationModesSection, {
-     t: t,
-     draft: draft,
-     setTop: setTop,
-     writable: writable,
+     t, draft, setTop, writable,
     }),
     // Voice Action Macros (#167)
     h(MacrosSection, {
-     t: t,
-     draft: draft,
-     setTop: setTop,
-     writable: writable,
+     t, draft, setTop, writable,
     }),
     h(JargonEditor, {
-     t: t,
-     draft: draft,
-     setTop: setTop,
-     writable: writable,
+     t, draft, setTop, writable,
      jargonInput: jargonInput,
      setJargonInput: setJargonInput,
     }),
@@ -5030,6 +4985,16 @@ function MacrosSection(props) {
    const isSecure = typeof window !== 'undefined' ? (window.isSecureContext || window.location.protocol === 'https:' || window.location.hostname === 'localhost' || window.location.hostname === '127.0.0.1') : true
    const hasMediaDevices = typeof navigator !== 'undefined' && !!navigator.mediaDevices && typeof navigator.mediaDevices.getUserMedia === 'function'
    const hasMediaRecorder = typeof MediaRecorder !== 'undefined'
+   const isDesktop = typeof navigator !== 'undefined' && (
+    /dsh-desktop|electron/i.test(navigator.userAgent || '') ||
+    (typeof window !== 'undefined' && window.location.protocol === 'dsh-app:')
+   )
+   const isWindows = typeof navigator !== 'undefined' && (
+    /win/i.test(navigator.platform || '') || /windows/i.test(navigator.userAgent || '')
+   )
+   const isMac = typeof navigator !== 'undefined' && (
+    /mac/i.test(navigator.platform || '') || /macintosh|mac os x/i.test(navigator.userAgent || '')
+   )
 
    const [micPermission, setMicPermission] = React.useState('prompt')
    const [devices, setDevices] = React.useState([])
@@ -5104,7 +5069,16 @@ function MacrosSection(props) {
    const constraints = {
    audio: selectedDev ? { deviceId: { exact: selectedDev } } : true,
    }
-   const stream = await navigator.mediaDevices.getUserMedia(constraints)
+   let stream
+   try {
+    stream = await navigator.mediaDevices.getUserMedia(constraints)
+   } catch (firstErr) {
+    if (selectedDev) {
+     stream = await navigator.mediaDevices.getUserMedia({ audio: true })
+    } else {
+     throw firstErr
+    }
+   }
    signalStreamRef.current = stream
    setMicPermission('granted')
    refreshDevices()
@@ -5157,7 +5131,15 @@ function MacrosSection(props) {
    const constraints = {
    audio: selectedDev ? { deviceId: { exact: selectedDev } } : true,
    }
-   testStream = await navigator.mediaDevices.getUserMedia(constraints)
+   try {
+    testStream = await navigator.mediaDevices.getUserMedia(constraints)
+   } catch (firstErr) {
+    if (selectedDev) {
+     testStream = await navigator.mediaDevices.getUserMedia({ audio: true })
+    } else {
+     throw firstErr
+    }
+   }
    let mime = 'audio/webm;codecs=opus'
    if (!MediaRecorder.isTypeSupported(mime)) mime = ''
    recorder = mime ? new MediaRecorder(testStream, { mimeType: mime }) : new MediaRecorder(testStream)
@@ -5241,10 +5223,19 @@ function MacrosSection(props) {
    }
 
    let nextStepHint = ''
+   const isPermDenied = micPermission === 'denied' || (testError && /permission|denied|notallowed/i.test(testError))
    if (!isSecure) {
   nextStepHint = t('wizardHintHttps')
-   } else if (micPermission === 'denied') {
-  nextStepHint = t('wizardHintPermission')
+   } else if (isPermDenied) {
+  if (isDesktop && isWindows) {
+   nextStepHint = t('wizardHintWinDesktop')
+  } else if (isDesktop && isMac) {
+   nextStepHint = t('wizardHintMacDesktop')
+  } else if (isWindows) {
+   nextStepHint = t('wizardHintWinBrowser')
+  } else {
+   nextStepHint = t('wizardHintPermission')
+  }
    } else if (peakRms === 0 && signalTesting) {
   nextStepHint = t('wizardHintDevice')
    } else if (testError) {
@@ -5300,6 +5291,13 @@ function MacrosSection(props) {
     className: 'cb-btn',
     onClick: requestPermissionAndTestSignal,
     }, t('wizardRequestMic')),
+    (isWindows && isPermDenied) ? h('a', {
+     href: 'ms-settings:privacy-microphone',
+     target: '_blank',
+     rel: 'noopener noreferrer',
+     className: 'cb-btn cb-btn-mini',
+     style: { marginLeft: '8px', textDecoration: 'none', display: 'inline-flex', alignItems: 'center' },
+    }, '⚙ ' + t('wizardOpenWinMicSettings')) : null,
    ),
    ),
   ),
@@ -5383,8 +5381,15 @@ function MacrosSection(props) {
   ),
 
   // Next Step Hint / Feedback
-  nextStepHint ? h('div', { className: 'dvs-wait', style: { fontStyle: 'italic' } },
-   t('wizardNextStepHint').replace('{hint}', nextStepHint)
+  nextStepHint ? h('div', { className: 'dvs-wait', style: { fontStyle: 'italic', display: 'flex', alignItems: 'center', flexWrap: 'wrap', gap: '8px' } },
+   h('span', null, t('wizardNextStepHint').replace('{hint}', nextStepHint)),
+   (isWindows && isPermDenied) ? h('a', {
+    href: 'ms-settings:privacy-microphone',
+    target: '_blank',
+    rel: 'noopener noreferrer',
+    className: 'cb-btn cb-btn-mini',
+    style: { textDecoration: 'none', fontStyle: 'normal' },
+   }, '⚙ ' + t('wizardOpenWinMicSettings')) : null,
   ) : null,
 
   // Wizard Bottom Actions
@@ -5484,9 +5489,7 @@ function MacrosSection(props) {
 
     exports.inject = ['timer', 'slots', 'configForms', 'locale']
     exports.apply = function apply(ctx) {
-      // English is the source and fallback language; Chinese is the built-in
-      // localized user dictionary. Other languages (like Russian) come from
-      // the separate translation plugin at runtime.
+      // English source/fallback, Chinese built-in dictionary.
       const addLocale = (locale, dictionary) => {
         try {
           return ctx.locale.register(NS, locale, dictionary)
@@ -5518,6 +5521,7 @@ function MacrosSection(props) {
       cycleDictationMode,
       resolveBrowserRecognitionLang,
       VoiceSection,
+      FirstRunWizard,
       cutPhrase,
       sendAudio,
       modeChain,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-voice",
-  "version": "0.9.44",
+  "version": "0.9.45",
   "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",