@goodandready/dsh-voice 0.8.31 → 0.8.32

package/README.md

@@ -84,12 +84,36 @@ graph LR
 * 📊 **Provider Latency & Health Dashboard**: Live visual telemetry of provider latency (ms), success rates, and errors directly within the settings UI.
 * 🔒 **Zero API Key Leakage**: Keys are resolved on the host via `ctx.credentials` (`credentialRef`) and never transmitted to browser clients.
 * 🖥️ **Offline Local Whisper Server**: Automatically boots and manages [whisper.cpp](https://github.com/ggerganov/whisper.cpp) (`whisper-server`) with on-the-fly `ffmpeg` transcode.
-* ⚡ **SenseVoice-ONNX / Sherpa-ONNX** *(0.8.11)*: Ultra-fast (~50–100ms) non-autoregressive local STT engine with automatic emotion/event tag stripping. Supports both Sherpa-ONNX HTTP and OpenAI-compatible endpoints.
+* ⚡ **SenseVoice-ONNX / Sherpa-ONNX** *(0.8.11, updated 0.8.32)*: Ultra-fast (~50–100ms) non-autoregressive local STT engine with 1-click automatic model installation to `~/.dsh/models/sensevoice`.
+* 🎙️ **Gated Turn-Taking & Echo Prevention** *(0.8.32)*: Automatically mutes the mic while the assistant is speaking (`dsh:tts:start`/`stop`) and supports speech-triggered Barge-In.
+* 👻 **Live Ghost Interim Preview** *(0.8.32)*: Visual real-time preview of spoken words before final chunk transcription.
+* ⏱️ **SVG Silence Ring Timer** *(0.8.32)*: Circular animated countdown indicator in the composer dock during the pending message delay.
+* 🗣️ **Hands-free Voice Actions** *(0.8.32)*: Spoken commands ("send", "cancel", "clear", "new line") trigger UI actions directly.
+* 🧩 **Structured Prompt Voice Selection** *(0.8.32)*: Spoken options automatically select buttons in interactive assistant choice prompts.
+* 📖 **Developer Lexicon & IT Jargon Correction** *(0.8.32)*: Phonetic normalization for developer slang (GitHub, Docker, Kubernetes, pnpm) with customizable settings dictionary.
 * 🌐 **Realtime Audio Streaming** *(0.8.11)*: Low-latency WebSocket bridge (`/dsh-voice/realtime`) for OpenAI Realtime API or local Sherpa-ONNX streaming. API keys stay securely on the host.
 * 🌊 **Liquid Wave & Dynamic Orb Visualizer** *(0.8.12)*: Smooth animated audio visualization in the recording pill with real-time mic volume reactivity. Switch between organic multi-layer liquid waves, pulsating radiant orb, classic bars, or off.
 
 ---
 
+## 📝 What's New in 0.8.32
+
+10-point feature epic & SenseVoice 1-Click Installer (Issue #129):
+
+| Area | Feature | Description |
+|------|---------|-------------|
+| Local ASR | **1-Click SenseVoice-Small** | Automated download and extraction of `model.int8.onnx` and `tokens.txt` directly to `~/.dsh/models/sensevoice` via host loopback endpoint. |
+| Turn-Taking | **Gated Mode** | Microphone input is muted when assistant audio starts (`dsh:tts:start`) and unmuted on `dsh:tts:stop`, eliminating acoustic feedback. |
+| Interruption | **Barge-In** | Speaking immediately signals assistant TTS to pause and abort current speech playback. |
+| Feedback | **Ghost Interim Preview** | Semi-transparent live text preview in composer pill during recognition before sentence finalizing. |
+| Countdown | **SVG Silence Ring** | Circular SVG progress ring visually ticking down pending auto-send delay. |
+| Control | **Hands-free Actions** | Spoken control words ("send", "cancel", "clear", "new line") trigger actions instead of becoming message text. |
+| Prompts | **Structured Prompts** | Spoken replies automatically match and submit options in active harness interactive prompts. |
+| Vocabulary | **IT Jargon Normalizer** | Auto-corrects spoken developer slang to canonical spelling (GitHub, Docker, Kubernetes, etc.) with custom dictionary in Settings. |
+| Configuration | **Individual Toggles** | Dedicated switches for every enhancement in Settings card with native `--dsw-alias-*` token styling. |
+
+---
+
 ## 📝 What's New in 0.8.19
 
 Quality batch after the v0.8.18 review (Gitea #79–#87, PR #88):

package/README.ru.md

@@ -84,12 +84,36 @@ graph LR
 * 📊 **Дашборд задержки и здоровья провайдеров**: мониторинг скорости ответа (мс), процента успешных транскрипций и ошибок в реальном времени.
 * 🔒 **Безопасность API-ключей**: ключи читаются на сервере через `ctx.credentials` и никогда не попадают в браузер клиента.
 * 🖥️ **Автозапуск локального whisper.cpp**: управление жизненным циклом `whisper-server` с авто-конвертацией через `ffmpeg`.
-* ⚡ **SenseVoice-ONNX / Sherpa-ONNX** *(0.8.11)*: сверхбыстрый (~50–100мс) неавторегрессивный локальный STT с автоматической очисткой тегов эмоций/событий. Поддержка Sherpa-ONNX HTTP и OpenAI-совместимых эндпоинтов.
+* ⚡ **SenseVoice-ONNX / Sherpa-ONNX** *(0.8.11, обновлено 0.8.32)*: сверхбыстрый (~50–100мс) неавторегрессивный локальный STT с установкой модели SenseVoice-Small в 1 клик в `~/.dsh/models/sensevoice`.
+* 🎙️ **Gated Turn-Taking и защита от эха** *(0.8.32)*: автоматическое выключение микрофона во время ответа ассистента (`dsh:tts:start`/`stop`) и поддержка прерывания речи (Barge-In).
+* 👻 **Живой предпросмотр (Ghost Interim Preview)** *(0.8.32)*: визуальное отображение распознаваемых слов в реальном времени до фиксации фрагмента.
+* ⏱️ **Кольцевой таймер тишины (SVG Silence Ring)** *(0.8.32)*: круговая анимация обратного отсчёта до авто-отправки в полоске композера.
+* 🗣️ **Голосовые команды управления (Hands-free Actions)** *(0.8.32)*: фразы «отправь», «отмена», «очисти», «с новой строки» выполняют действия напрямую, не попадая в текст сообщения.
+* 🧩 **Голосовой выбор в интерактивах (Structured Prompts)** *(0.8.32)*: произнесение вариантов ответов нажимает соответствующие кнопки в активных модальных окнах ассистента.
+* 📖 **Словарь IT-жаргона и сленга разработчиков** *(0.8.32)*: фонетическое исправление сленга (гитхаб -> GitHub, кубер -> Kubernetes, пнпм -> pnpm) с возможностью добавления своих правил в настройках.
 * 🌐 **Потоковое аудио в реальном времени** *(0.8.11)*: низколатентный WebSocket-мост (`/dsh-voice/realtime`) для OpenAI Realtime API или локального Sherpa-ONNX. API-ключи остаются на хосте.
 * 🌊 **Анимированные визуализаторы Liquid Wave и Dynamic Orb** *(0.8.12)*: живая интерактивная анимация звуковой волны в полоске записи с откликом на громкость микрофона. Выбор стиля: текучая волна, пульсирующая сфера, классические столбики или выключен.
 
 ---
 
+## 📝 Что нового в 0.8.32
+
+Пакет из 10 улучшений и установка SenseVoice в 1 клик (Issue #129):
+
+| Область | Функция | Описание |
+|---------|---------|----------|
+| Локальный STT | **Установка SenseVoice в 1 клик** | Автоматическая загрузка и распаковка `model.int8.onnx` и `tokens.txt` в `~/.dsh/models/sensevoice` через защищённый loopback-маршрут. |
+| Координация речи | **Gated Turn-Taking** | Микрофон глушится при воспроизведении ответа ассистента (`dsh:tts:start`) и включается обратно при `dsh:tts:stop`, исключая эхо и зацикливание. |
+| Прерывание | **Barge-In** | Начало речи пользователя мгновенно прерывает воспроизведение ответа синтезатора речи. |
+| Обратная связь | **Ghost-предпросмотр** | Полупрозрачный текст живого распознавания прямо в пилюле композера до окончания фразы. |
+| Обратный отсчёт | **SVG Silence Ring** | Круговой векторный индикатор оставшегося времени до автоматической отправки сообщения. |
+| Управление | **Голосовые действия** | Команды «отправь», «отмена», «очисти» выполняют команды UI без вставки текста в поле ввода. |
+| Интерактивы | **Структурированные промпты** | Голосовой выбор вариантов ответов во всплывающих вопросах ассистента. |
+| Словарь | **Коррекция IT-жаргона** | Автоматическая замена сленга программистов на правильное написание с редактором правил в настройках. |
+| Конфигурация | **Индивидуальные переключатели** | Каждая функция отдельно включается и выключается в карточке настроек. |
+
+---
+
 ## 📝 Что нового в 0.8.19
 
 Quality-батч после ревью v0.8.18 (Gitea #79–#87, PR #88):

package/lib/client.js

@@ -204,6 +204,27 @@ window.__ModuleLoader__.load({
       'updating': 'Updating…',
       'updateSuccess': 'Updated successfully! Restart DSH to apply.',
       'updateFailed': 'Update failed. Check logs.',
+      'sensevoiceTitle': 'SenseVoice Local ASR (1-Click Setup)',
+      'sensevoiceDesc': 'Ultra-fast offline speech recognition on CPU (50–100ms latency, Russian/English).',
+      'installSensevoice': 'Install SenseVoice-Small (1-Click)',
+      'installingSensevoice': 'Installing SenseVoice…',
+      'sensevoiceReady': 'Ready',
+      'sensevoiceNotInstalled': 'Not Installed',
+      'sensevoiceDownloading': 'Downloading model…',
+      'turnTakingTitle': 'Turn-Taking & Audio Coordination',
+      'turnTakingDesc': 'Coordinate speaking and listening so assistant audio does not feed back into the microphone.',
+      'gatedTurnTaking': 'Gated Mode (mute mic while assistant is speaking)',
+      'bargeInDesc': 'True Barge-In (speaking interrupts assistant speech)',
+      'smartActionsTitle': 'Hands-free & Smart Actions',
+      'autoSendVisualRing': 'Silence Ring countdown animation',
+      'voiceActionsLabel': 'Voice action commands ("send", "cancel", "clear", "new line")',
+      'liveInterimPreview': 'Live speech preview (ghost text preview)',
+      'structuredPromptVoice': 'Spoken answers for interactive choice prompts',
+      'jargonTitle': 'Developer Lexicon & IT Jargon',
+      'techJargonCorrection': 'Auto-correct technical jargon (GitHub, Docker, pnpm, Kubernetes, etc.)',
+      'jargonAdd': 'Add replacement',
+      'jargonFrom': 'Spoken slang',
+      'jargonTo': 'Correct spelling',
     }
 
     const zh = {
@@ -376,6 +397,27 @@ window.__ModuleLoader__.load({
       'updating': '正在更新…',
       'updateSuccess': '更新成功！请重启 DSH 生效。',
       'updateFailed': '更新失败，请查看日志。',
+      'sensevoiceTitle': 'SenseVoice 本地 ASR (一键部署)',
+      'sensevoiceDesc': 'CPU 上超低延迟离线语音识别 (50–100ms，中英俄多语种)。',
+      'installSensevoice': '一键安装 SenseVoice-Small',
+      'installingSensevoice': '正在安装 SenseVoice…',
+      'sensevoiceReady': '已就绪',
+      'sensevoiceNotInstalled': '未安装',
+      'sensevoiceDownloading': '正在下载模型…',
+      'turnTakingTitle': '轮流对话与音频协同',
+      'turnTakingDesc': '协调听与说，避免扬声器播放的声音被麦克风二次拾音。',
+      'gatedTurnTaking': '门控模式 (助手播报时麦克风自动静音)',
+      'bargeInDesc': '实时打断 (说话时立即打断助手朗读)',
+      'smartActionsTitle': '免提与智能指令',
+      'autoSendVisualRing': '环形倒计时动画指示器',
+      'voiceActionsLabel': '语音动作指令 ("发送", "取消", "清空", "换行")',
+      'liveInterimPreview': '实时语音文本预览 (幽灵文本)',
+      'structuredPromptVoice': '语音应答交互式选择提示',
+      'jargonTitle': '开发者词汇与 IT 术语',
+      'techJargonCorrection': '自动校正开发术语 (GitHub, Docker, pnpm, Kubernetes 等)',
+      'jargonAdd': '添加替换项',
+      'jargonFrom': '口语词汇',
+      'jargonTo': '标准拼写',
     }
 
     // Strings are also needed outside components — in recording handlers and
@@ -416,7 +458,17 @@ window.__ModuleLoader__.load({
       '.dvo-mic-test{display:flex;align-items:center;gap:10px;margin-top:8px;padding:6px 12px;background:var(--dsw-alias-bg-layer-2);border-radius:8px;border:1px solid var(--dsw-alias-border-l1)}' +
       '.dvo-meter-bar{flex:1;height:8px;background:var(--dsw-alias-bg-layer-3);border-radius:4px;overflow:hidden;position:relative}' +
       '.dvo-meter-fill{height:100%;background:var(--dsw-alias-state-success-primary);border-radius:4px;transition:width 60ms ease-out}' +
-      '.dvo-meter-gate{position:absolute;top:0;bottom:0;width:2px;background:var(--dsw-alias-state-warning-primary);box-shadow:0 0 4px var(--dsw-alias-state-warning-primary);z-index:2;pointer-events:none}'
+      '.dvo-meter-gate{position:absolute;top:0;bottom:0;width:2px;background:var(--dsw-alias-state-warning-primary);box-shadow:0 0 4px var(--dsw-alias-state-warning-primary);z-index:2;pointer-events:none}' +
+      '.dvo-silence-ring{width:34px;height:34px;transform:rotate(-90deg);flex:none}' +
+      '.dvo-ring-bg{fill:none;stroke:var(--dsw-alias-border-l1);stroke-width:2.5}' +
+      '.dvo-ring-fg{fill:none;stroke:var(--dsw-alias-state-info-primary);stroke-width:2.5;stroke-linecap:round;transition:stroke-dashoffset 100ms linear}' +
+      '.dvo-ghost{font-style:italic;color:var(--dsw-alias-label-tertiary);margin-left:6px;max-width:240px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;font-size:12px}' +
+      '.dvo-installer-box{display:flex;flex-direction:column;gap:8px;padding:10px 12px;background:var(--dsw-alias-bg-layer-2);border:1px solid var(--dsw-alias-border-l1);border-radius:8px;margin-top:8px}' +
+      '.dvo-installer-row{display:flex;align-items:center;justify-content:space-between;gap:8px}' +
+      '.dvo-progress-bar{height:6px;width:100%;background:var(--dsw-alias-bg-layer-3);border-radius:3px;overflow:hidden}' +
+      '.dvo-progress-fill{height:100%;background:var(--dsw-alias-state-info-primary);transition:width 150ms ease}' +
+      '.dvo-jargon-table{width:100%;margin-top:6px;border-collapse:collapse;font-size:12px}' +
+      '.dvo-jargon-table td{padding:4px 6px;border-bottom:1px solid var(--dsw-alias-border-l1)}'
     const cssId = 'dsh-voice/client.module.css'
     if (typeof document !== 'undefined' && !document.querySelector('style[data-dsh-plugin="dsh-voice"][data-plugin-css="' + cssId + '"]')) {
       const tag = document.createElement('style')
@@ -435,15 +487,88 @@ window.__ModuleLoader__.load({
       pending: null,      // {text, leftMs} — message-mode cancel window
       lastNote: null,     // {blob, url, mime, text} — last recording
       showPlayer: false,
+      interimTranscript: '', // real-time live preview (ghost text)
+      isTtsSpeaking: false,  // turn-taking: true while assistant speech is active
+      actionTriggered: null, // 'send' | 'cancel' | 'clear' | 'newline' | null
       inputActions: null,
       input: null,
-      settings: { vadSilenceMs: 700, autoSendMs: 4000, stream: false, streamChunkMs: 1200, vadAdapt: 0, noiseSuppression: true, contextGlossary: true },
+      settings: {
+        vadSilenceMs: 700, autoSendMs: 4000, stream: false, streamChunkMs: 1200, vadAdapt: 0,
+        noiseSuppression: true, contextGlossary: true,
+        gatedTurnTaking: true, autoSendVisualRing: true, voiceActions: true,
+        techJargonCorrection: true, jargonDictionary: [], liveInterimPreview: true,
+        structuredPromptVoice: true,
+      },
       listeners: new Set(),
       notify() { this.listeners.forEach((l) => l()) },
       set(patch) { Object.assign(this, patch); this.notify() },
       subscribe(l) { this.listeners.add(l); return () => this.listeners.delete(l) },
     }
 
+    if (typeof window !== 'undefined') {
+      window.addEventListener('dsh:tts:start', () => {
+        voice.set({ isTtsSpeaking: true })
+      })
+      window.addEventListener('dsh:tts:stop', () => {
+        voice.set({ isTtsSpeaking: false })
+      })
+    }
+
+    function triggerBargeIn() {
+      if (typeof window === 'undefined') return
+      try {
+        window.dispatchEvent(new CustomEvent('dsh:tts:cancel', { bubbles: true }))
+        window.dispatchEvent(new CustomEvent('dsh:tts:stop', { bubbles: true }))
+        const audios = document.querySelectorAll('audio')
+        for (const a of audios) {
+          if (!a.paused) { try { a.pause() } catch {} }
+        }
+      } catch { /* best effort */ }
+      voice.set({ isTtsSpeaking: false })
+    }
+
+    function findActiveStructuredPrompt() {
+      if (typeof document === 'undefined') return null
+      try {
+        const modal = document.querySelector('.dsh-structured-question, [data-dsh-dialog="question"], .dialog-prompt, .cb-modal-question')
+        if (modal) {
+          const input = modal.querySelector('textarea, input[type="text"]')
+          const buttons = Array.from(modal.querySelectorAll('button:not([disabled])'))
+          return { modal, input, buttons }
+        }
+      } catch { /* best effort */ }
+      return null
+    }
+
+    function submitToStructuredPrompt(text, promptInfo) {
+      if (!promptInfo) return false
+      const trimmed = String(text || '').trim().toLowerCase()
+      const { buttons, input } = promptInfo
+      if (buttons && buttons.length > 0) {
+        if (/^(1|первый|first)$/i.test(trimmed) && buttons[0]) { buttons[0].click(); return true }
+        if (/^(2|второй|second)$/i.test(trimmed) && buttons[1]) { buttons[1].click(); return true }
+        if (/^(3|третий|third)$/i.test(trimmed) && buttons[2]) { buttons[2].click(); return true }
+        if (/^(да|yes|ok|confirm|согласен)$/i.test(trimmed)) {
+          const yesBtn = buttons.find((b) => /да|yes|ok|confirm/i.test(b.textContent || ''))
+          if (yesBtn) { yesBtn.click(); return true }
+        }
+        if (/^(нет|no|отмена|cancel)$/i.test(trimmed)) {
+          const noBtn = buttons.find((b) => /нет|no|cancel|отмена/i.test(b.textContent || ''))
+          if (noBtn) { noBtn.click(); return true }
+        }
+      }
+      if (input) {
+        input.value = text
+        input.dispatchEvent(new Event('input', { bubbles: true }))
+        const form = input.closest('form')
+        if (form) {
+          form.dispatchEvent(new Event('submit', { bubbles: true, cancelable: true }))
+        }
+        return true
+      }
+      return false
+    }
+
     function useVoice() {
       const [, force] = React.useReducer((x) => x + 1, 0)
       React.useEffect(() => voice.subscribe(force), [])
@@ -909,6 +1034,37 @@ window.__ModuleLoader__.load({
           const out = await sendAudio(blob, rec.mime, 'dictation')
           if (rec.cancelled) return
           const text = out && out.text ? out.text : ''
+          const action = out && out.action ? out.action : null
+
+          if (voice.settings.structuredPromptVoice !== false && text) {
+            const promptInfo = findActiveStructuredPrompt()
+            if (promptInfo && submitToStructuredPrompt(text, promptInfo)) {
+              return
+            }
+          }
+
+          if (action === 'send') {
+            if (text) appendDraft(text)
+            if (voice.inputActions && typeof voice.inputActions.send === 'function') {
+              voice.inputActions.send()
+            }
+            return
+          }
+          if (action === 'cancel') {
+            cancelCurrent()
+            return
+          }
+          if (action === 'clear') {
+            if (voice.inputActions && typeof voice.inputActions.clear === 'function') {
+              voice.inputActions.clear()
+            }
+            return
+          }
+          if (action === 'newline') {
+            appendDraft(text ? (text + '\n') : '\n')
+            return
+          }
+
           const delay = Number(voice.settings.sendDelayMs) || 0
           if (text && delay > 0 && !voice.holding) {
             // Delayed insert with a cancel window (#29-5): text is already
@@ -937,7 +1093,11 @@ window.__ModuleLoader__.load({
         lang: language,
         continuous: true,
         onInterim: (text) => {
-          voice.caption = text; voice.notify()
+          voice.caption = text
+          if (voice.settings.liveInterimPreview !== false) {
+            voice.interimTranscript = text
+          }
+          voice.notify()
           // Wake-word (#45): if interim starts with the trigger phrase, stop
           // browser listening and switch to normal recording.
           const ww = String(voice.settings.wakeWord || '').trim().toLowerCase()
@@ -946,6 +1106,7 @@ window.__ModuleLoader__.load({
             if (t.startsWith(ww)) {
               voice.browser = null
               voice.caption = ''
+              voice.interimTranscript = ''
               openMic(mode)
                 .then((rec) => {
                   if (voice.phase !== 'recording') {
@@ -962,6 +1123,7 @@ window.__ModuleLoader__.load({
         onFinal: (text) => {
           finals.push(text)
           voice.caption = ''
+          voice.interimTranscript = ''
           // Dictation appends immediately; messages accumulate until release.
           if (mode === 'dictation') appendDraft(text)
           else voice.notify()
@@ -969,6 +1131,7 @@ window.__ModuleLoader__.load({
         onError: (reason) => {
           // Browser failed after start — say so instead of pretending to listen.
           voice.browser = null
+          voice.interimTranscript = ''
           voice.set({ phase: 'error', error: t('browserFailed') + reason })
         },
       })
@@ -979,11 +1142,17 @@ window.__ModuleLoader__.load({
 
     function startRecording(mode) {
       if (voice.phase !== 'idle' && voice.phase !== 'error') return
-      // Announce before opening the mic: the sooner playback mutes, the less
-      // of it ends up in the recording.
+      if (voice.isTtsSpeaking) {
+        if (voice.settings.bargeIn) {
+          triggerBargeIn()
+        } else if (voice.settings.gatedTurnTaking !== false) {
+          voice.set({ phase: 'error', error: 'Assistant speaking (gated mode)' })
+          return
+        }
+      }
       announceVoice('start')
       dictationQueue = Promise.resolve()
-      voice.set({ phase: 'recording', mode, error: '', levels: [], caption: '' })
+      voice.set({ phase: 'recording', mode, error: '', levels: [], caption: '', interimTranscript: '' })
       modeChain(mode).then((info) => {
         // Browser leg only when it is explicitly first in the chain.
         if (info.chain[0] === 'browser' && startBrowserLeg(mode, info.language)) return
@@ -1138,11 +1307,38 @@ window.__ModuleLoader__.load({
           const out = await sendAudio(blob, rec.mime, mode)
           if (out.command) { runSessionCommand(out.command); return }
           const text = out.text || ''
-          if (!text) { voice.set({ phase: 'error', error: t('nothingHeard') }); return }
+          const action = out.action || null
+          if (!text && !action) { voice.set({ phase: 'error', error: t('nothingHeard') }); return }
+
+          if (voice.settings.structuredPromptVoice !== false && text) {
+            const promptInfo = findActiveStructuredPrompt()
+            if (promptInfo && submitToStructuredPrompt(text, promptInfo)) {
+              voice.set({ phase: 'idle' })
+              return
+            }
+          }
+
+          if (action === 'send') {
+            if (text) appendDraft(text)
+            submitPending()
+            return
+          }
+          if (action === 'cancel') {
+            cancelCurrent()
+            return
+          }
+          if (action === 'clear') {
+            if (voice.inputActions && typeof voice.inputActions.clear === 'function') {
+              voice.inputActions.clear()
+            }
+            voice.set({ phase: 'idle' })
+            return
+          }
+
           appendDraft(text)
           if (voice.lastNote) voice.lastNote.text = text
           if (mode === 'message') {
-            voice.set({ phase: 'pending', pending: { text: text, leftMs: voice.settings.autoSendMs, audioUrl: noteUrl } })
+            voice.set({ phase: 'pending', pending: { text: text, totalMs: voice.settings.autoSendMs, leftMs: voice.settings.autoSendMs, audioUrl: noteUrl } })
           } else {
             voice.set({ phase: 'idle' })
           }
@@ -1577,7 +1773,10 @@ window.__ModuleLoader__.load({
                 ? t('holdHint')
                 : (inBrowser
                   ? t('listening')
-                  : (rec && rec.hadSpeech ? (v.speaking ? t('speaking') : t('silence')) : hint)))),
+                  : (rec && rec.hadSpeech ? (v.speaking ? t('speaking') : t('silence')) : hint))),
+            v.interimTranscript && v.settings.liveInterimPreview !== false
+              ? React.createElement('span', { className: 'dvo-ghost' }, v.interimTranscript)
+              : null),
           React.createElement('button', {
             type: 'button', className: 'dvo-pbtn', title: t('stop'),
             'aria-label': t('stop'),
@@ -1631,6 +1830,23 @@ window.__ModuleLoader__.load({
                 React.createElement('span', { className: 'dvo-audio-time' }, t('listenBack')),
               )
             : null,
+          v.settings.autoSendVisualRing !== false
+            ? (() => {
+                const total = (voice.pending && voice.pending.totalMs) || Number(voice.settings.autoSendMs) || 4000
+                const leftMs = (voice.pending && voice.pending.leftMs) || 0
+                const frac = total > 0 ? Math.max(0, Math.min(1, leftMs / total)) : 0
+                const circ = 2 * Math.PI * 14
+                const off = circ * (1 - frac)
+                return React.createElement('svg', { className: 'dvo-silence-ring', viewBox: '0 0 34 34' },
+                  React.createElement('circle', { cx: 17, cy: 17, r: 14, className: 'dvo-ring-bg' }),
+                  React.createElement('circle', {
+                    cx: 17, cy: 17, r: 14, className: 'dvo-ring-fg',
+                    strokeDasharray: `${circ} ${circ}`,
+                    strokeDashoffset: off,
+                  }),
+                )
+              })()
+            : null,
           React.createElement('span', { className: 'dvo-status', 'aria-live': 'polite', 'aria-atomic': 'true' }, t('sendingIn')),
           React.createElement('span', { className: 'dvo-count' }, left + t('secondsShort')),
           React.createElement('button', {
@@ -1693,6 +1909,13 @@ window.__ModuleLoader__.load({
                 noiseGateDb: Number(data && data.noiseGateDb !== undefined ? data.noiseGateDb : -45),
                 contextGlossary: data && data.contextGlossary !== false,
                 visualizerStyle: (data && data.visualizerStyle) || 'liquid-wave',
+                gatedTurnTaking: data && data.gatedTurnTaking !== false,
+                autoSendVisualRing: data && data.autoSendVisualRing !== false,
+                voiceActions: data && data.voiceActions !== false,
+                techJargonCorrection: data && data.techJargonCorrection !== false,
+                jargonDictionary: Array.isArray(data && data.jargonDictionary) ? data.jargonDictionary : [],
+                liveInterimPreview: data && data.liveInterimPreview !== false,
+                structuredPromptVoice: data && data.structuredPromptVoice !== false,
               })
             })
             .catch(() => { /* no host hint — no hotkey */ })
@@ -2028,9 +2251,57 @@ window.__ModuleLoader__.load({
           noiseGateDb: Number(val.noiseGateDb !== undefined ? val.noiseGateDb : -45),
           contextGlossary: val.contextGlossary !== false,
           visualizerStyle: val.visualizerStyle || 'liquid-wave',
+          gatedTurnTaking: val.gatedTurnTaking !== false,
+          autoSendVisualRing: val.autoSendVisualRing !== false,
+          voiceActions: val.voiceActions !== false,
+          techJargonCorrection: val.techJargonCorrection !== false,
+          liveInterimPreview: val.liveInterimPreview !== false,
+          structuredPromptVoice: val.structuredPromptVoice !== false,
         })
       }, [ready, value])
 
+      const [sensevoiceState, setSensevoiceState] = React.useState(null)
+      const [installingSensevoice, setInstallingSensevoice] = React.useState(false)
+
+      const fetchSensevoice = () => {
+        fetch('/dsh-voice/sensevoice-installer')
+          .then((r) => r.json())
+          .then((d) => {
+            setSensevoiceState(d)
+            setInstallingSensevoice(!!(d && d.installing))
+          })
+          .catch(() => {})
+      }
+
+      React.useEffect(() => {
+        fetchSensevoice()
+      }, [])
+
+      const triggerInstallSensevoice = () => {
+        setInstallingSensevoice(true)
+        fetch('/dsh-voice/sensevoice-installer', {
+          method: 'POST',
+          headers: { 'x-dsh-plugin-update': '1' },
+        })
+          .then((r) => r.json())
+          .then(() => {
+            const timer = setInterval(() => {
+              fetch('/dsh-voice/sensevoice-installer')
+                .then((r) => r.json())
+                .then((d) => {
+                  setSensevoiceState(d)
+                  if (!d.installing) {
+                    clearInterval(timer)
+                    setInstallingSensevoice(false)
+                    fetchStatus()
+                  }
+                })
+                .catch(() => clearInterval(timer))
+            }, 2000)
+          })
+          .catch(() => setInstallingSensevoice(false))
+      }
+
       // Interactive microphone tester
       const [testingMic, setTestingMic] = React.useState(false)
       const [testLevel, setTestLevel] = React.useState(0)
@@ -2553,7 +2824,159 @@ window.__ModuleLoader__.load({
           ),
         ),
 
-        // 6. Provider Telemetry & Latency Dashboard
+        // 6. SenseVoice Local ASR (1-Click Setup)
+        React.createElement('div', { className: 'cb-section-card' },
+          React.createElement('div', { className: 'cb-section-title' },
+            React.createElement('span', null, '⚡ ' + t('sensevoiceTitle')),
+            sensevoiceState && sensevoiceState.installed
+              ? React.createElement('span', { className: 'cb-badge cb-badge-ok' }, t('sensevoiceReady') + ' (small-int8)')
+              : React.createElement('span', { className: 'cb-badge cb-badge-warn' },
+                  installingSensevoice ? t('sensevoiceDownloading') : t('sensevoiceNotInstalled')
+                ),
+          ),
+          React.createElement('div', { className: 'cb-section-desc' }, t('sensevoiceDesc')),
+          React.createElement('div', { className: 'dvo-installer-box' },
+            React.createElement('div', { className: 'cb-row', style: { alignItems: 'center', justifyContent: 'space-between', gap: '12px' } },
+              React.createElement('div', { style: { fontSize: '12px', color: 'var(--dsw-alias-content-secondary)' } },
+                sensevoiceState && sensevoiceState.path
+                  ? ('~/' + sensevoiceState.path.replace(/^.*[\\/](\.dsh[\\/]models[\\/].*)$/, '$1'))
+                  : '~/.dsh/models/sensevoice'
+              ),
+              React.createElement('button', {
+                type: 'button',
+                className: 'cb-btn cb-btn-primary',
+                style: { fontSize: '12px', padding: '6px 12px' },
+                disabled: installingSensevoice || !writable,
+                onClick: triggerInstallSensevoice,
+              }, installingSensevoice ? ('⏳ ' + t('installingSensevoice')) : ('📥 ' + t('installSensevoice'))),
+            ),
+            installingSensevoice
+              ? React.createElement('div', { style: { marginTop: '8px' } },
+                  React.createElement('div', { className: 'dvo-progress-bar' },
+                    React.createElement('div', { className: 'dvo-progress-bar-fill' })
+                  ),
+                  React.createElement('div', { style: { fontSize: '11px', marginTop: '4px', color: 'var(--dsw-alias-content-secondary)' } },
+                    (sensevoiceState && sensevoiceState.progress) || t('sensevoiceDownloading')
+                  )
+                )
+              : null
+          )
+        ),
+
+        // 7. Turn-Taking & Audio Coordination
+        React.createElement('div', { className: 'cb-section-card' },
+          React.createElement('div', { className: 'cb-section-title' },
+            React.createElement('span', null, '🎙️ ' + t('turnTakingTitle')),
+          ),
+          React.createElement('div', { className: 'cb-section-desc' }, t('turnTakingDesc')),
+          React.createElement('div', { className: 'cb-grid-2' },
+            React.createElement('label', { className: 'cb-check-label' },
+              React.createElement('input', {
+                type: 'checkbox',
+                checked: draft ? draft.gatedTurnTaking !== false : true,
+                disabled: !writable,
+                onChange: (e) => setTop('gatedTurnTaking', e.target.checked),
+              }),
+              t('gatedTurnTaking'),
+            ),
+            React.createElement('label', { className: 'cb-check-label' },
+              React.createElement('input', {
+                type: 'checkbox',
+                checked: !!(draft && draft.bargeIn),
+                disabled: !writable,
+                onChange: (e) => setTop('bargeIn', e.target.checked),
+              }),
+              t('bargeInDesc'),
+            ),
+          ),
+        ),
+
+        // 8. Hands-free & Smart Actions
+        React.createElement('div', { className: 'cb-section-card' },
+          React.createElement('div', { className: 'cb-section-title' },
+            React.createElement('span', null, '✨ ' + t('smartActionsTitle')),
+          ),
+          React.createElement('div', { className: 'cb-grid-2' },
+            React.createElement('label', { className: 'cb-check-label' },
+              React.createElement('input', {
+                type: 'checkbox',
+                checked: draft ? draft.liveInterimPreview !== false : true,
+                disabled: !writable,
+                onChange: (e) => setTop('liveInterimPreview', e.target.checked),
+              }),
+              t('liveInterimPreview'),
+            ),
+            React.createElement('label', { className: 'cb-check-label' },
+              React.createElement('input', {
+                type: 'checkbox',
+                checked: draft ? draft.autoSendVisualRing !== false : true,
+                disabled: !writable,
+                onChange: (e) => setTop('autoSendVisualRing', e.target.checked),
+              }),
+              t('autoSendVisualRing'),
+            ),
+            React.createElement('label', { className: 'cb-check-label' },
+              React.createElement('input', {
+                type: 'checkbox',
+                checked: draft ? draft.voiceActions !== false : true,
+                disabled: !writable,
+                onChange: (e) => setTop('voiceActions', e.target.checked),
+              }),
+              t('voiceActionsLabel'),
+            ),
+            React.createElement('label', { className: 'cb-check-label' },
+              React.createElement('input', {
+                type: 'checkbox',
+                checked: draft ? draft.structuredPromptVoice !== false : true,
+                disabled: !writable,
+                onChange: (e) => setTop('structuredPromptVoice', e.target.checked),
+              }),
+              t('structuredPromptVoice'),
+            ),
+          ),
+        ),
+
+        // 9. Developer Lexicon & IT Jargon
+        React.createElement('div', { className: 'cb-section-card' },
+          React.createElement('div', { className: 'cb-section-title' },
+            React.createElement('span', null, '📖 ' + t('jargonTitle')),
+          ),
+          React.createElement('label', { className: 'cb-check-label', style: { marginBottom: '8px' } },
+            React.createElement('input', {
+              type: 'checkbox',
+              checked: draft ? draft.techJargonCorrection !== false : true,
+              disabled: !writable,
+              onChange: (e) => setTop('techJargonCorrection', e.target.checked),
+            }),
+            t('techJargonCorrection'),
+          ),
+          React.createElement('div', { className: 'cb-field' },
+            React.createElement('label', null, t('jargonAdd') + ' (spoken -> replacement)'),
+            React.createElement('textarea', {
+              rows: 3,
+              disabled: !writable,
+              placeholder: 'кубик -> k8s\nворктри -> worktree',
+              value: Array.isArray(draft && draft.jargonDictionary)
+                ? draft.jargonDictionary.map((item) => item && item.from ? `${item.from} -> ${item.to}` : '').filter(Boolean).join('\n')
+                : '',
+              onChange: (e) => {
+                const lines = e.target.value.split('\n')
+                const parsed = lines
+                  .map((line) => {
+                    const parts = line.split(/->|=|:/)
+                    if (parts.length >= 2) {
+                      return { from: parts[0].trim(), to: parts.slice(1).join(':').trim() }
+                    }
+                    return null
+                  })
+                  .filter(Boolean)
+                setTop('jargonDictionary', parsed)
+              },
+            }),
+          ),
+        ),
+
+        // 10. Provider Telemetry & Latency Dashboard
         React.createElement('div', { className: 'cb-section-card' },
           React.createElement('div', { className: 'cb-section-title' },
             React.createElement('span', null, '📊 ' + t('providerDashboard')),
@@ -2615,7 +3038,7 @@ window.__ModuleLoader__.load({
           React.createElement('div', { className: 'cb-section-desc' }, t('versionHint')),
           React.createElement('div', { className: 'cb-row', style: { alignItems: 'center', gap: '12px' } },
             React.createElement('span', { className: 'cb-badge cb-badge-ok' },
-              'v' + ((updaterStatus && updaterStatus.currentVersion) || '0.8.29')
+              'v' + ((updaterStatus && updaterStatus.currentVersion) || '0.8.32')
             ),
             updaterStatus && updaterStatus.updateAvailable
               ? React.createElement('span', { className: 'cb-badge cb-badge-warn' },

package/lib/index.js

@@ -20,13 +20,14 @@ import path from 'node:path'
 import { runChain } from './chain.js'
 import { makeProviders, PROVIDER_KEYS, PRESET_KEYS, KNOWN_KEYS, DEFAULT_MODELS, CUSTOM_TEMPLATES } from './providers.js'
 import { toWav16k } from './wav.js'
-import { normalizePhrase } from './normalize.js'
+import { normalizePhrase, applyJargonDictionary, extractVoiceActions } from './normalize.js'
 import { createStatsTracker, mergeContextVocabulary } from './stats.js'
 import { writeJson, readBody, isTrustedCaller } from './http-util.js'
 import { createPolishText } from './polish.js'
 import { createUserMessage } from '@deepseek-ai/dsh-llm'
 import { buildProviderOrder, sessionCommand } from './transcribe-core.js'
 import { registerPluginUpdater } from './updater.js'
+import { registerSensevoiceInstaller } from './sensevoice-installer.js'
 
 function isAutoLang(lang) {
   return !lang || lang === 'auto' || String(lang).includes(',')
@@ -164,6 +165,20 @@ export const Config = z.object({
     .description('Start the local SenseVoice / sherpa-onnx server process automatically on startup.'),
   visualizerStyle: z.union(['liquid-wave', 'dynamic-orb', 'bars', 'off']).default('liquid-wave')
     .description('Audio visualizer animation style in the recording pill: "liquid-wave", "dynamic-orb", "bars", or "off".'),
+  gatedTurnTaking: z.boolean().default(true)
+    .description('Gated mode for speakers: pause recording or ignore input while assistant TTS is speaking to prevent acoustic feedback.'),
+  autoSendVisualRing: z.boolean().default(true)
+    .description('Show animated circular countdown ring on recording pill before auto-sending.'),
+  voiceActions: z.boolean().default(true)
+    .description('Enable spoken action triggers ("send", "cancel", "clear", "new line" in English/Russian).'),
+  techJargonCorrection: z.boolean().default(true)
+    .description('Apply developer lexicon and IT jargon normalization (GitHub, Docker, pnpm, Kubernetes, etc.).'),
+  jargonDictionary: z.array(z.object({ from: z.string(), to: z.string() })).default([])
+    .description('Custom jargon replacement dictionary.'),
+  liveInterimPreview: z.boolean().default(true)
+    .description('Display real-time interim speech preview (ghost text) in the recording pill.'),
+  structuredPromptVoice: z.boolean().default(true)
+    .description('Target voice dictation and numeric choices to active DSH structured prompt modals.'),
 })
 
 const MIME_BY_EXT = {
@@ -179,13 +194,14 @@ export function apply(ctx, baseConfig) {
   // host registers it via settings.register the snapshot is empty and
   // read-only. Reading through live() means edits apply to the next request
   // instead of requiring a process restart.
+  let scope = null
   let getConfig = () => baseConfig
   const live = () => Config(structuredClone(getConfig() ?? {})) ?? baseConfig
 
   ctx.inject(['settings'], (sctx) => {
-    const scope = sctx.settings.register(NS, Config, { base: baseConfig })
+    scope = sctx.settings.register(NS, Config, { base: baseConfig })
     getConfig = () => scope.get() ?? baseConfig
-    sctx.effect(() => () => { getConfig = () => baseConfig })
+    sctx.effect(() => () => { getConfig = () => baseConfig; scope = null })
   })
 
   async function resolveKey(ref) {
@@ -397,6 +413,13 @@ export function apply(ctx, baseConfig) {
         sensevoiceAutostart: !!cfg.sensevoiceAutostart,
         deepgramBaseUrl: cfg.deepgramBaseUrl || 'https://api.deepgram.com',
         visualizerStyle: cfg.visualizerStyle || 'liquid-wave',
+        gatedTurnTaking: cfg.gatedTurnTaking !== false,
+        autoSendVisualRing: cfg.autoSendVisualRing !== false,
+        voiceActions: cfg.voiceActions !== false,
+        techJargonCorrection: cfg.techJargonCorrection !== false,
+        jargonDictionary: Array.isArray(cfg.jargonDictionary) ? cfg.jargonDictionary : [],
+        liveInterimPreview: cfg.liveInterimPreview !== false,
+        structuredPromptVoice: cfg.structuredPromptVoice !== false,
         providerStats: statsTracker.get(),
       })
     },
@@ -476,8 +499,19 @@ export function apply(ctx, baseConfig) {
           const cmd = sessionCommand(out.text)
           if (cmd) { writeJson(res, 200, { ok: true, command: cmd, provider: out.provider, tookMs: out.tookMs }); return }
         }
-        const text = await polishText(out.text, modeCfg, controller.signal)
-        writeJson(res, 200, { ok: true, text, provider: out.provider, tookMs: out.tookMs })
+        let text = await polishText(out.text, modeCfg, controller.signal)
+        if (cfg.techJargonCorrection !== false) {
+          text = applyJargonDictionary(text, cfg.jargonDictionary)
+        }
+        let action = null
+        if (cfg.voiceActions !== false) {
+          const actionRes = extractVoiceActions(text)
+          if (actionRes.action) {
+            text = actionRes.text
+            action = actionRes.action
+          }
+        }
+        writeJson(res, 200, { ok: true, text, action, provider: out.provider, tookMs: out.tookMs })
       } catch (e) {
         writeJson(res, 502, { ok: false, error: { code: 'chain', message: String(e && e.message || e) } })
       } finally {
@@ -549,4 +583,16 @@ export function apply(ctx, baseConfig) {
     packageName: '@goodandready/dsh-voice',
     manifestUrl: new URL('../package.json', import.meta.url),
   }), 'dsh-voice: updater route')
+
+  ctx.effect(() => registerSensevoiceInstaller(ctx, {
+    liveConfig: live,
+    isAlive: sensevoiceAlive,
+    updateConfig: async (patch) => {
+      if (scope && typeof scope.patch === 'function') {
+        scope.patch(patch)
+      }
+    },
+    startSensevoice,
+  }), 'dsh-voice: sensevoice installer route')
 }
+

package/lib/normalize.js

@@ -47,11 +47,99 @@ export function wordsToDigits(text) {
   })
 }
 
+export const DEFAULT_JARGON_DICTIONARY = [
+  { from: 'гитхаб', to: 'GitHub' },
+  { from: 'гитхабе', to: 'GitHub' },
+  { from: 'гитхаба', to: 'GitHub' },
+  { from: 'гитлаб', to: 'GitLab' },
+  { from: 'гитлабе', to: 'GitLab' },
+  { from: 'гитлаба', to: 'GitLab' },
+  { from: 'дш', to: 'DSH' },
+  { from: 'пнпм', to: 'pnpm' },
+  { from: 'нпм', to: 'npm' },
+  { from: 'докер', to: 'Docker' },
+  { from: 'докера', to: 'Docker' },
+  { from: 'докером', to: 'Docker' },
+  { from: 'докер-контейнер', to: 'Docker-контейнер' },
+  { from: 'докер-контейнеры', to: 'Docker-контейнеры' },
+  { from: 'кубер', to: 'Kubernetes' },
+  { from: 'кубернетис', to: 'Kubernetes' },
+  { from: 'кубернетес', to: 'Kubernetes' },
+  { from: 'тайпскрипт', to: 'TypeScript' },
+  { from: 'тайпскрипте', to: 'TypeScript' },
+  { from: 'тайпскрипта', to: 'TypeScript' },
+  { from: 'яваскрипт', to: 'JavaScript' },
+  { from: 'джаваскрипт', to: 'JavaScript' },
+  { from: 'яваскрипта', to: 'JavaScript' },
+  { from: 'джаваскрипта', to: 'JavaScript' },
+  { from: 'нода', to: 'Node.js' },
+  { from: 'ноде', to: 'Node.js' },
+  { from: 'ноды', to: 'Node.js' },
+  { from: 'пулл реквест', to: 'PR' },
+  { from: 'пулл-реквест', to: 'PR' },
+  { from: 'пайплайн', to: 'pipeline' },
+  { from: 'пайплайна', to: 'pipeline' },
+  { from: 'коммит', to: 'commit' },
+  { from: 'коммита', to: 'commit' },
+  { from: 'коммиты', to: 'commits' },
+  { from: 'бранч', to: 'branch' },
+  { from: 'бренч', to: 'branch' },
+  { from: 'ворктри', to: 'worktree' },
+]
+
+export function applyJargonDictionary(text, customReplacements = []) {
+  if (!text || typeof text !== 'string') return text || ''
+  const dict = [...(customReplacements || []), ...DEFAULT_JARGON_DICTIONARY]
+  let result = text
+  for (const item of dict) {
+    if (!item?.from || !item?.to) continue
+    const escaped = item.from.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')
+    const regex = new RegExp(`(?<![a-zA-Zа-яА-ЯёЁ0-9_-])${escaped}(?![a-zA-Zа-яА-ЯёЁ0-9_-])`, 'gi')
+    result = result.replace(regex, item.to)
+  }
+  return result
+}
+
+export function extractVoiceActions(text) {
+  const raw = String(text || '').trim()
+  if (!raw) return { text: '', action: null }
+
+  const lower = raw.toLowerCase().replace(/[.!?…]+$/, '').trim()
+  if (/^(отправь|отправить|send|send it|send message)$/i.test(lower)) {
+    return { text: '', action: 'send' }
+  }
+  if (/^(отмена|cancel|discard)$/i.test(lower)) {
+    return { text: '', action: 'cancel' }
+  }
+  if (/^(очисти|очистить|clear|clear composer)$/i.test(lower)) {
+    return { text: '', action: 'clear' }
+  }
+  if (/^(новая строка|с новой строки|new line|newline)$/i.test(lower)) {
+    return { text: '', action: 'newline' }
+  }
+
+  const trailingSend = /[\s,]+(отправь|отправить|send|send it)[.!?…]*$/i
+  if (trailingSend.test(raw)) {
+    return { text: raw.replace(trailingSend, '').trim(), action: 'send' }
+  }
+  const trailingCancel = /[\s,]+(отмена|cancel)[.!?…]*$/i
+  if (trailingCancel.test(raw)) {
+    return { text: raw.replace(trailingCancel, '').trim(), action: 'cancel' }
+  }
+  const trailingNewline = /[\s,]+(новая строка|с новой строки|new line)[.!?…]*$/i
+  if (trailingNewline.test(raw)) {
+    return { text: raw.replace(trailingNewline, '').trim() + '\n', action: 'newline' }
+  }
+
+  return { text: raw, action: null }
+}
+
 export function normalizePhrase(text, opts = {}) {
   let out = String(text || '').trim()
   if (!out) return out
   if (opts.digits) out = wordsToDigits(out)
   if (opts.commaSpacing) out = fixCommaSpacing(out)
+  if (opts.jargon) out = applyJargonDictionary(out, opts.customJargon)
   if (opts.capSentences) out = capitalizeSentences(out)
   if (opts.trailingPeriod) out = ensureTrailingPeriod(out)
   return out

package/lib/sensevoice-installer.js

@@ -0,0 +1,227 @@
+// lib/sensevoice-installer.js — 1-Click installer for SenseVoice-Small / sherpa-onnx
+// Downloads pre-quantized ONNX model and configures autostart on the host.
+
+import { createWriteStream, existsSync, readdirSync } from 'node:fs'
+import { mkdir, stat } from 'node:fs/promises'
+import { homedir } from 'node:os'
+import path from 'node:path'
+import { spawn } from 'node:child_process'
+import { writeJson, isTrustedCaller } from './http-util.js'
+
+export const SENSEVOICE_MODEL_URL =
+  'https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2'
+export const SENSEVOICE_MODEL_MIRROR =
+  'https://huggingface.co/csukuangfj/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/resolve/main/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2'
+
+let installerState = {
+  installing: false,
+  progress: 0,
+  step: 'idle',
+  error: null,
+  modelPath: '',
+}
+
+export function getDefaultModelDir() {
+  return path.join(homedir(), '.dsh', 'models', 'sensevoice')
+}
+
+export function findModelFile(dir) {
+  if (!dir || !existsSync(dir)) return null
+  try {
+    const entries = readdirSync(dir, { recursive: true, withFileTypes: true })
+    for (const entry of entries) {
+      if (entry.isFile() && (entry.name === 'model.int8.onnx' || entry.name === 'model.onnx')) {
+        const parent = entry.parentPath || (entry.path ? entry.path : dir)
+        return path.join(parent, entry.name)
+      }
+    }
+  } catch { /* directory read failed */ }
+  return null
+}
+
+export function findTokensFile(dir) {
+  if (!dir || !existsSync(dir)) return null
+  try {
+    const entries = readdirSync(dir, { recursive: true, withFileTypes: true })
+    for (const entry of entries) {
+      if (entry.isFile() && entry.name === 'tokens.txt') {
+        const parent = entry.parentPath || (entry.path ? entry.path : dir)
+        return path.join(parent, entry.name)
+      }
+    }
+  } catch { /* directory read failed */ }
+  return null
+}
+
+export async function getSensevoiceStatus(cfg, isAliveFn) {
+  const modelDir = cfg?.sensevoiceModel ? path.dirname(cfg.sensevoiceModel) : getDefaultModelDir()
+  const modelFile = cfg?.sensevoiceModel && existsSync(cfg.sensevoiceModel)
+    ? cfg.sensevoiceModel
+    : findModelFile(getDefaultModelDir())
+  const tokensFile = modelFile ? findTokensFile(path.dirname(modelFile)) : null
+
+  const running = typeof isAliveFn === 'function' ? await isAliveFn() : false
+  const installed = Boolean(modelFile && existsSync(modelFile))
+
+  return {
+    installed,
+    running,
+    modelPath: modelFile || '',
+    tokensPath: tokensFile || '',
+    installing: installerState.installing,
+    progress: installerState.progress,
+    step: installerState.step,
+    error: installerState.error,
+  }
+}
+
+export async function downloadArchive(url, destPath, onProgress) {
+  const res = await fetch(url, { signal: AbortSignal.timeout(600_000) })
+  if (!res.ok) throw new Error(`Download failed with HTTP ${res.status}`)
+
+  const totalBytes = Number(res.headers.get('content-length')) || 150_000_000
+  let receivedBytes = 0
+
+  const fileStream = createWriteStream(destPath)
+  const reader = res.body.getReader()
+
+  while (true) {
+    const { done, value } = await reader.read()
+    if (done) break
+    fileStream.write(value)
+    receivedBytes += value.length
+    if (typeof onProgress === 'function') {
+      const pct = Math.min(95, Math.round((receivedBytes / totalBytes) * 90))
+      onProgress(pct, `downloading (${Math.round(receivedBytes / 1024 / 1024)} MB)`)
+    }
+  }
+
+  await new Promise((resolve, reject) => {
+    fileStream.end()
+    fileStream.on('finish', resolve)
+    fileStream.on('error', reject)
+  })
+}
+
+export function extractTarBz2(archivePath, targetDir) {
+  return new Promise((resolve, reject) => {
+    const child = spawn('tar', ['-xjf', archivePath, '-C', targetDir])
+    let stderr = ''
+    child.stderr.on('data', (d) => { stderr += d })
+    child.on('close', (code) => {
+      if (code === 0) resolve()
+      else reject(new Error(`tar extraction failed with code ${code}: ${stderr}`))
+    })
+    child.on('error', reject)
+  })
+}
+
+export async function startSensevoiceInstall(ctx, cfg, updateConfigFn, startSensevoiceFn) {
+  if (installerState.installing) {
+    throw new Error('SenseVoice installation is already in progress')
+  }
+
+  installerState = {
+    installing: true,
+    progress: 5,
+    step: 'preparing',
+    error: null,
+    modelPath: '',
+  }
+
+  const modelDir = getDefaultModelDir()
+  const archivePath = path.join(modelDir, 'sensevoice.tar.bz2')
+
+  try {
+    await mkdir(modelDir, { recursive: true })
+    installerState.step = 'downloading'
+    installerState.progress = 10
+
+    let downloadOk = false
+    try {
+      await downloadArchive(SENSEVOICE_MODEL_URL, archivePath, (pct, status) => {
+        installerState.progress = pct
+        installerState.step = status
+      })
+      downloadOk = true
+    } catch (e1) {
+      // Fallback to mirror
+      installerState.step = 'downloading (mirror)'
+      await downloadArchive(SENSEVOICE_MODEL_MIRROR, archivePath, (pct, status) => {
+        installerState.progress = pct
+        installerState.step = status
+      })
+      downloadOk = true
+    }
+
+    if (!downloadOk) throw new Error('Could not download SenseVoice model from primary or mirror URL')
+
+    installerState.step = 'extracting'
+    installerState.progress = 95
+    await extractTarBz2(archivePath, modelDir)
+
+    const modelFile = findModelFile(modelDir)
+    if (!modelFile) throw new Error('model.int8.onnx not found after extraction')
+
+    installerState.modelPath = modelFile
+    installerState.progress = 100
+    installerState.step = 'ready'
+    installerState.installing = false
+
+    if (typeof updateConfigFn === 'function') {
+      await updateConfigFn({
+        sensevoiceModel: modelFile,
+        sensevoiceAutostart: true,
+      })
+    }
+
+    if (typeof startSensevoiceFn === 'function') {
+      await startSensevoiceFn()
+    }
+
+    return { ok: true, modelPath: modelFile }
+  } catch (err) {
+    installerState.installing = false
+    installerState.error = String(err?.message || err)
+    installerState.step = 'failed'
+    throw err
+  }
+}
+
+export function registerSensevoiceInstaller(ctx, { liveConfig, isAlive, updateConfig, startSensevoice }) {
+  if (!ctx.webServer || typeof ctx.webServer.register !== 'function') return
+
+  ctx.webServer.register({
+    kind: 'exact',
+    path: '/dsh-voice/sensevoice-installer',
+    handler: async (req, res) => {
+      if (req.method === 'GET') {
+        const status = await getSensevoiceStatus(liveConfig(), isAlive)
+        writeJson(res, 200, status)
+        return
+      }
+
+      if (req.method === 'POST') {
+        if (!isTrustedCaller(req)) {
+          writeJson(res, 403, { ok: false, error: 'Forbidden: untrusted caller' })
+          return
+        }
+
+        if (installerState.installing) {
+          writeJson(res, 409, { ok: false, error: 'Installation already in progress', state: installerState })
+          return
+        }
+
+        // Start background install
+        startSensevoiceInstall(ctx, liveConfig(), updateConfig, startSensevoice).catch((err) => {
+          ctx.logger?.warn?.(`SenseVoice 1-click install error: ${String(err)}`)
+        })
+
+        writeJson(res, 202, { ok: true, message: 'SenseVoice installation started', state: installerState })
+        return
+      }
+
+      writeJson(res, 405, { ok: false, error: 'Method not allowed' })
+    },
+  })
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-voice",
-  "version": "0.8.31",
+  "version": "0.8.32",
   "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",