@goodandready/dsh-tts 0.4.28 → 0.4.29

package/CHANGELOG.md

@@ -1,5 +1,34 @@
 # Changelog
 
+## 0.4.29
+
+### Fixed
+- **Settings & Revision Management** (#151, #152):
+  - Added robust `extractRevision` supporting nested Schemastery namespaces and entries array.
+  - Return revision in `GET /dsh-tts/config` and send it in `PUT /dsh-tts/config`.
+  - Eliminated stale `savedConfig` caching in `live()`, enabling direct live Schemastery volatile box evaluation.
+- **Audio Pipeline, Tool Schema & Dedup** (#156, #157, #166, #173, #175, #190):
+  - Enforced speech preparation, maxChars, and summarizeReply before sentence splitting in `speakAsItGoes`.
+  - Added `id: { type: 'string' }` to `speak_text` tool output schema under `additionalProperties: false` preventing core ToolOutputError.
+  - Included credential slot reference in cache and flight fingerprints avoiding cross-credential collisions.
+  - In-memory index for synthesis cache avoiding O(N) fs scans.
+- **Stream Hub & Web Player** (#158, #159, #160, #161, #174, #176, #186):
+  - `skipCurrent()` now cancels active browser SpeechSynthesis and clears busy state.
+  - `playAudio` resolves promise on NotAllowedError and detaches gesture listeners cleanly.
+  - Disconnect slow SSE subscribers exceeding backpressure buffer threshold.
+  - Registered updater routes inside ctx.effect with clean disposers.
+- **Providers & Local Engines** (#165, #167, #168, #169, #170, #171, #172, #177, #178, #179, #180):
+  - Nested integrations parser in UI and gated /speak on disabled channels.
+  - Canonical EN defaults for Edge and eSpeak.
+  - 100% key parity for EN/ZH model hints.
+  - Google TTS validates MIME and wraps raw PCM in valid WAV container.
+  - Azure SSML includes required xmlns/mstts attributes.
+  - Robust model download locking, cancellation awaiting writer cleanup, and HTML error page rejection.
+- **Release Hygiene & Sanitization** (#183, #184, #185):
+  - Removed outdated marketing claims from READMEs.
+  - Restored internal DEV docs while verifying exclusion from npm pack and GitHub mirror.
+  - Cleaned up publish-github script to handle negation patterns.
+
 ## 0.4.28
 
 ### Fixed

package/README.md

@@ -2,7 +2,7 @@
 
 <div align="center">
 
-<h3>Multi-Provider Text-to-Speech Voice Synthesis with Local Neural Engines, Sub-300ms Streaming, IT Dictionary & Messenger Integration for DeepSeek Harness</h3>
+<h3>Multi-Provider Text-to-Speech Voice Synthesis with Local Neural Engines, Real-time Streaming, IT Dictionary & Messenger Integration for DeepSeek Harness</h3>
 
 <p align="center">
   <a href="https://www.npmjs.com/package/@goodandready/dsh-tts"><img src="https://img.shields.io/npm/v/@goodandready/dsh-tts.svg?style=for-the-badge&color=6366f1&labelColor=1e1b4b" alt="npm version"></a>
@@ -49,7 +49,7 @@ graph LR
 
     subgraph Stream [Low-Latency Streaming]
         Scrub --> SSE[SSE /dsh-tts/stream]
-        SSE --> Worklet[AudioWorklet PCM Processor]
+        SSE --> Player[Web Audio & HTML Audio Player]
     end
 
     subgraph Cache [Performance Layer]
@@ -59,7 +59,7 @@ graph LR
 
     subgraph Fallback [TTS Provider Fallback Chain]
         LRU -->|Cache Miss| Chain{Active Chain}
-        Chain -->|Offline| P1[Edge TTS / Piper / eSpeak]
+        Chain -->|Online / Local| P1[Edge TTS / Piper / eSpeak]
         Chain -.->|Cloud Neural| P2[OpenAI / ElevenLabs / Google / Azure / Groq]
         Chain -.->|OpenAI-compatible| P3[SiliconFlow / DeepInfra / Fireworks / OpenRouter]
         Chain -.->|Other| P4[MiMo / MiniMax / Custom]
@@ -85,13 +85,14 @@ graph LR
 
 ## 🚀 Key Features
 
-### 1. 📴 Offline system engines + optional future neural runtimes
-* **Edge TTS / Piper / eSpeak**: fully offline or free local/system synthesis without cloud API keys (Edge needs the `edge-tts` CLI).
+### 1. 📴 Local and system engines + optional future neural runtimes
+* **Piper / eSpeak**: fully offline local synthesis without cloud API keys.
+* **Edge TTS**: free online cloud synthesis without API keys (requires network connection and the `edge-tts` CLI).
 * **Kokoro-82M / F5-TTS**: weight download and status UI only. **Neural inference is not bundled** in this package — those providers fail with a clear reason and the fallback chain continues. Do not enable them expecting speech until a supported runtime is wired.
 * **ModelManager UI**: Direct manual installation in settings with real-time download progress bar, SHA-256 validation, and deletion. No silent or automatic multi-gigabyte downloads.
 
 ### 2. ⚡ Real-Time Streaming Audio (< 300 ms Latency)
-* **AudioWorklet (`TTSWorklet`)**: High-performance Web Audio Worklet processor playing seamless Float32Array PCM chunks at 24 kHz without audible clicks or buffer underruns.
+* **HTML Audio & Web Audio Player**: Seamless queued audio playback with prebuffering, barge-in support, playback rate control, and instant browser fallback.
 * **Server-Sent Events (SSE)**: Dedicated `/dsh-tts/stream` route delivering synthesized chunks to connected browsers instantly.
 
 ### 3. 🎙️ Voice Duplex & VAD Barge-In (with `@goodandready/dsh-voice`)
@@ -173,7 +174,7 @@ dsh plugin --profile web add @goodandready/dsh-tts
 
 ---
 
-## ⚙️ Configuration Recipes (`settings.yaml`)
+## ⚙️ Configuration Recipes (DSH Settings UI)
 
 ```yaml
 dsh-tts:
@@ -212,7 +213,7 @@ The plugin settings card (**Settings → Plugins → Plugin settings**) exposes
 
 Provider **API keys are never stored in plugin settings**. Paste them in the chain editor; values go to the DSH credential store via `PUT /dsh-tts/credential`.
 
-Config-only (not in the card): `*KeyEnv` fields (`openaiKeyEnv`, `elevenlabsKeyEnv`, …). They only rename the credential slot the plugin looks up. Change them in `settings.yaml` if you must rebind a key name; the defaults match the usual environment variable names.
+Config-only (not in the card): `*KeyEnv` fields (`openaiKeyEnv`, `elevenlabsKeyEnv`, …). They only rename the credential slot the plugin looks up. Change them in the profile settings if you must rebind a key name; the defaults match the usual environment variable names.
 
 
 ## 🤖 HTTP Endpoints Reference

package/README.ru.md

@@ -2,7 +2,7 @@
 
 <div align="center">
 
-<h3>Многопровайдерная озвучка ответов агента: облачные и системные offline-движки, потоковый звук, IT-словарь и интеграция с мессенджерами для DeepSeek Harness</h3>
+<h3>Многопровайдерная озвучка ответов агента: облачные, локальные и системные движки, потоковый звук, IT-словарь и интеграция с мессенджерами для DeepSeek Harness</h3>
 
 <p align="center">
   <a href="https://www.npmjs.com/package/@goodandready/dsh-tts"><img src="https://img.shields.io/npm/v/@goodandready/dsh-tts.svg?style=for-the-badge&color=6366f1&labelColor=1e1b4b" alt="npm version"></a>
@@ -37,11 +37,11 @@
 
 ## ⚡ Обзор
 
-> **Честно про локальные нейросети:** Kokoro/F5 — только статус и загрузка весов. Inference **не входит** в пакет; эти провайдеры падают с понятной причиной, цепочка идёт дальше. Рабочий offline: Edge TTS, Piper, eSpeak.
+> **Честно про локальные нейросети:** Kokoro/F5 — только статус и загрузка весов. Inference **не входит** в пакет; эти провайдеры падают с понятной причиной, цепочка идёт дальше. Рабочий offline: Piper, eSpeak (Edge TTS — бесплатный онлайн-сервис без ключей).
 
 **`dsh-tts`** обеспечивает качественное голосовое озвучивание ответов ассистента в веб-интерфейсе **DeepSeek Harness**. При включённой опции **«Читать ответы агента»**, каждая готовая реплика или потоковый фрагмент синтезируется на хосте и мгновенно воспроизводится в браузере.
 
-API-ключи никогда не передаются в браузер: синтез аудио выполняется на стороне хоста через **независимые цепочки отказоустойчивости (фолбеков)**, включая системные offline-движки (Edge TTS, Piper, eSpeak). Kokoro/F5 — только загрузка весов; inference не входит в пакет.
+API-ключи никогда не передаются в браузер: синтез аудио выполняется на стороне хоста через **независимые цепочки отказоустойчивости (фолбеков)**, включая локальные и системные движки (Edge TTS — онлайн без ключей, Piper и eSpeak — offline). Kokoro/F5 — только загрузка весов; inference не входит в пакет.
 
 ```mermaid
 graph LR
@@ -51,7 +51,7 @@ graph LR
 
     subgraph Stream [Потоковый звук < 300 мс]
         Scrub --> SSE[SSE /dsh-tts/stream]
-        SSE --> Worklet[AudioWorklet PCM процессор]
+        SSE --> Player[Web Audio & HTML Audio плеер]
     end
 
     subgraph Cache [Слой производительности]
@@ -61,7 +61,7 @@ graph LR
 
     subgraph Fallback [Цепочка фолбеков TTS]
         LRU -->|Промах кэша| Chain{Активная цепочка}
-        Chain -->|Offline| P1[Edge TTS / Piper / eSpeak]
+        Chain -->|Online / Local| P1[Edge TTS / Piper / eSpeak]
         Chain -.->|Cloud| P2[OpenAI / ElevenLabs / Google / Azure / Groq]
         Chain -.->|Бесплатные облака| P3[EdgeTTS / SiliconFlow]
         Chain -.->|Системный фолбек| P4[Локальный Piper / eSpeak NG]
@@ -87,13 +87,14 @@ graph LR
 
 ## 🚀 Ключевые возможности
 
-### 1. 📴 Offline-системные движки и честный статус нейросетей
-* **Edge TTS / Piper / eSpeak**: рабочий offline/системный синтез без облачных API-ключей.
+### 1. 📴 Локальные движки и честный статус нейросетей
+* **Piper / eSpeak**: полностью автономный локальный оффлайн-синтез.
+* **Edge TTS**: бесплатный онлайн-синтез через Microsoft без API-ключей (требует интернет и CLI `edge-tts`).
 * **Kokoro-82M / F5-TTS**: только UI и загрузка весов. **Inference не входит в пакет** — провайдер падает с понятной причиной, цепочка идёт дальше.
 * **ModelManager в UI**: Ручная загрузка моделей прямо из настроек плагина с отображением процентов и прогресс-бара, проверкой SHA-256 и кнопкой удаления. Никаких скрытых или автозагрузок гигабайтных весов.
 
 ### 2. ⚡ Потоковое воспроизведение звука с задержкой < 300 мс
-* **AudioWorklet (`TTSWorklet`)**: Высокопроизводительный процессор веб-аудио с бесшовным кольцевым буфером для воспроизведения PCM-чанков (24 кГц) без щелчков и пауз.
+* **HTML Audio & Web Audio плеер**: Бесшовное воспроизведение аудио-очереди с предбуферизацией, поддержкой перебивания речи (barge-in), регулировкой скорости и отказоустойчивым fallback на синтез браузера.
 * **Server-Sent Events (SSE)**: Роут `/dsh-tts/stream` передаёт синтезированные куски речи сразу в браузер, исключая задержки традиционного полинга.
 
 ### 3. 🎙️ Голосовой диалог и VAD Barge-In (в связке с `@goodandready/dsh-voice`)
@@ -156,7 +157,7 @@ dsh plugin --profile web add @goodandready/dsh-tts
 
 ---
 
-## ⚙️ Пример конфигурации (`settings.yaml`)
+## ⚙️ Пример конфигурации (Интерфейс настроек DSH)
 
 ```yaml
 dsh-tts:
@@ -195,7 +196,7 @@ dsh-tts:
 
 **API-ключи не хранятся в настройках плагина.** Вставляйте их в редакторе цепочки — значение уходит в credential store DSH через `PUT /dsh-tts/credential`.
 
-Только в конфиге (нет в карточке): поля `*KeyEnv` (`openaiKeyEnv`, `elevenlabsKeyEnv`, …). Они задают **имя** credential-слота; по умолчанию совпадают с привычными именами env-переменных. Меняйте в `settings.yaml`, если нужно перевязать имя ключа.
+Только в конфиге (нет в карточке): поля `*KeyEnv` (`openaiKeyEnv`, `elevenlabsKeyEnv`, …). Они задают **имя** credential-слота; по умолчанию совпадают с привычными именами env-переменных. Меняйте в настройках профиля, если нужно перевязать имя ключа.
 
 
 ## 🤖 HTTP роуты API

package/README.zh.md

@@ -47,9 +47,9 @@ graph LR
         Reply[💬 智能体生成文本] --> Scrub[智能过滤 & IT术语词典]
     end
 
-    subgraph Stream [低延迟流式音频 < 300ms]
+    subgraph Stream [实时流式音频播放]
         Scrub --> SSE[SSE /dsh-tts/stream]
-        SSE --> Worklet[AudioWorklet PCM 处理器]
+        SSE --> Player[Web Audio & HTML Audio 播放器]
     end
 
     subgraph Cache [性能缓存层]
@@ -90,8 +90,8 @@ graph LR
 * **Kokoro-82M / F5-TTS**：仅权重下载与状态展示。**推理未捆绑在本包中**，provider 会明确失败并继续 fallback。
 * **ModelManager 管理界面**：在设置中手动安装模型，实时显示下载进度条、SHA-256 校验和一键删除。无任何静默或自动下载。
 
-### 2. ⚡ 实时流式音频播放（延迟 < 300ms）
-* **AudioWorklet (`TTSWorklet`)**：高性能 Web Audio Worklet 处理器，以 24kHz 采样率无缝播放 Float32Array PCM 数据块，无可感知的爆音或缓冲区欠载。
+### 2. ⚡ 实时流式音频播放
+* **HTML Audio & Web Audio 播放器**：无缝队列音频播放，支持预缓冲、麦克风打断响应 (barge-in)、播放倍速控制以及浏览器原生语音故障回退。
 * **Server-Sent Events (SSE)**：专用 `/dsh-tts/stream` 路由将合成音频片段即时推送至浏览器。
 
 ### 3. 🎙️ 语音双工对话与 VAD 打断（配合 `@goodandready/dsh-voice`）
@@ -165,7 +165,7 @@ dsh plugin --profile web add @goodandready/dsh-tts
 API 密钥不写入插件设置，只在链编辑器粘贴，经 `PUT /dsh-tts/credential` 写入 DSH credential store。
 
 
-## ⚙️ 配置示例（`settings.yaml`）
+## ⚙️ 配置说明（DSH 设置面板）
 
 ```yaml
 dsh-tts:

package/lib/client.js

@@ -51,6 +51,14 @@ window.__ModuleLoader__.load({
     }
 
 
+        function getModelHint(provider, t) {
+      if (typeof t === 'function') {
+        const loc = t('modelHint_' + provider)
+        if (loc && !loc.startsWith('modelHint_')) return loc
+      }
+      return MODEL_HINT[provider] || ''
+    }
+
     function createErrorBoundary() {
       if (!React || typeof React.Component !== 'function') {
         return function NoopBoundary(props) { return (props && props.children) || null }
@@ -74,7 +82,7 @@ window.__ModuleLoader__.load({
                 className: 'dts-alert dts-alert-err',
                 style: { margin: '12px 0', padding: '14px', borderRadius: '8px' },
               },
-              React.createElement('div', { style: { fontWeight: 600, marginBottom: '6px' } }, '⚠️ TTS UI Error:'),
+              React.createElement('div', { style: { fontWeight: 600, marginBottom: '6px' } }, (typeof fallbackDockText === 'function' ? fallbackDockText('ttsUiError') : '⚠️ TTS UI Error:')),
               React.createElement('div', { style: { fontSize: '12px', wordBreak: 'break-all' } }, String((this.state.error && this.state.error.message) || this.state.error)),
               React.createElement(
                 'button',
@@ -165,7 +173,16 @@ window.__ModuleLoader__.load({
       }
     }
 
+        let clearGestureListeners = null
+    function removeGestureListeners() {
+      if (typeof clearGestureListeners === 'function') {
+        clearGestureListeners()
+        clearGestureListeners = null
+      }
+    }
+
     function unlockAudio() {
+      removeGestureListeners()
       try {
         const AC = window.AudioContext || window.webkitAudioContext
         if (AC) {
@@ -330,16 +347,21 @@ window.__ModuleLoader__.load({
               player.busy = false
               playerChanged()
               if (typeof window !== 'undefined') {
+                removeGestureListeners()
                 const onUserGesture = () => {
-                  window.removeEventListener('click', onUserGesture)
-                  window.removeEventListener('keydown', onUserGesture)
-                  window.removeEventListener('touchstart', onUserGesture)
+                  removeGestureListeners()
                   unlockAudio()
                 }
                 window.addEventListener('click', onUserGesture, { once: true })
                 window.addEventListener('keydown', onUserGesture, { once: true })
                 window.addEventListener('touchstart', onUserGesture, { once: true })
+                clearGestureListeners = () => {
+                  window.removeEventListener('click', onUserGesture)
+                  window.removeEventListener('keydown', onUserGesture)
+                  window.removeEventListener('touchstart', onUserGesture)
+                }
               }
+              resolve()
               return
             }
             done()
@@ -370,6 +392,9 @@ window.__ModuleLoader__.load({
     }
 
     function stopPlayback() {
+      removeGestureListeners()
+      player.blocked = false
+      setAutoplayBlocked(false)
       for (const item of player.queue) {
         if (item && item.prepared && item.prepared.url) {
           try { URL.revokeObjectURL(item.prepared.url) } catch { /* already revoked */ }
@@ -402,16 +427,22 @@ window.__ModuleLoader__.load({
     }
 
     function skipCurrent() {
+      try {
+        if (typeof window !== 'undefined' && window.speechSynthesis) {
+          window.speechSynthesis.cancel()
+        }
+      } catch (_err) { /* no speechSynthesis */ }
       if (player.audio) {
         try {
           player.audio.pause()
           player.audio.currentTime = 0
-        } catch { /* ignore pause error */ }
+        } catch (_err) { /* ignore pause error */ }
       }
       if (currentAudioResolve) {
         currentAudioResolve()
       } else {
         player.audio = null
+        player.busy = false
         playerChanged()
         if (player.queue.length) drainQueue()
       }
@@ -469,13 +500,13 @@ window.__ModuleLoader__.load({
           if (player.enabled === false) return
           player.queue.push(item)
           drainQueue()
-        } catch {}
+        } catch (_err) { /* ignore malformed chime payload */ }
       }
 
       const connect = () => {
         if (sseDisposed) return
         if (activeSse) {
-          try { activeSse.close() } catch {}
+          try { activeSse.close() } catch (_err) { /* active sse already closed */ }
           activeSse = null
         }
         try {
@@ -486,7 +517,7 @@ window.__ModuleLoader__.load({
           es.addEventListener('chime', onChime)
           es.onerror = () => {
             if (activeSse === es) {
-              try { es.close() } catch {}
+              try { es.close() } catch (_err) { /* eventsource already closed */ }
               activeSse = null
             }
             if (sseDisposed) return
@@ -504,7 +535,7 @@ window.__ModuleLoader__.load({
         sseDisposed = true
         if (activeSseTimer) clearTimeout(activeSseTimer)
         if (activeSse) {
-          try { activeSse.close() } catch {}
+          try { activeSse.close() } catch (_err) { /* active sse already closed on dispose */ }
           activeSse = null
         }
       }
@@ -655,6 +686,25 @@ window.__ModuleLoader__.load({
       return (map && map[provider]) || { configured: false, writable: true, ref: '' }
     }
     const en = {
+      'ttsUiError': '⚠️ TTS UI Error:',
+      'modelHint_openai': 'provider default',
+      'modelHint_elevenlabs': 'provider default',
+      'modelHint_google': 'provider default',
+      'modelHint_azure': 'region voice name',
+      'modelHint_groq': 'provider default',
+      'modelHint_deepgram': 'provider default (English-only voices)',
+      'modelHint_openrouter': 'provider/model id',
+      'modelHint_siliconflow': 'model id (e.g. FunAudioLLM/CosyVoice2-0.5B)',
+      'modelHint_deepinfra': 'model id (e.g. hexgrad/Kokoro-82M)',
+      'modelHint_fireworks': 'model id (e.g. kokoro)',
+      'modelHint_mimo': 'model id (e.g. mimo-v2.5-tts)',
+      'modelHint_custom': 'custom model',
+      'modelHint_minimax': 'model id',
+      'modelHint_edge': 'BCP-47 voice, e.g. en-US-AriaNeural',
+      'modelHint_piper': 'path to .onnx',
+      'modelHint_espeak': 'language code, e.g. en',
+      'modelHint_kokoro': 'not bundled — offline neural runtime unavailable',
+      'modelHint_f5': 'not bundled — offline neural runtime unavailable',
       'dockLabel': 'Speech',
       'dockPause': 'Pause',
       'dockResume': 'Resume speech',
@@ -862,6 +912,25 @@ window.__ModuleLoader__.load({
     }
 
     const zh = {
+      'ttsUiError': '⚠️ TTS 界面错误：',
+      'modelHint_openai': '服务商默认',
+      'modelHint_elevenlabs': '服务商默认',
+      'modelHint_google': '服务商默认',
+      'modelHint_azure': '区域语音名称',
+      'modelHint_groq': '服务商默认',
+      'modelHint_deepgram': '服务商默认 (仅英语语音)',
+      'modelHint_openrouter': '服务商/模型 ID',
+      'modelHint_siliconflow': '模型 ID (如 FunAudioLLM/CosyVoice2-0.5B)',
+      'modelHint_deepinfra': '模型 ID (如 hexgrad/Kokoro-82M)',
+      'modelHint_fireworks': '模型 ID (如 kokoro)',
+      'modelHint_mimo': '模型 ID (如 mimo-v2.5-tts)',
+      'modelHint_custom': '自定义模型',
+      'modelHint_minimax': '模型 ID',
+      'modelHint_edge': 'BCP-47 语音名称，如 en-US-AriaNeural',
+      'modelHint_piper': '.onnx 模型文件路径',
+      'modelHint_espeak': '语言代码，如 en',
+      'modelHint_kokoro': '未内置 — 离线神经语音运行时不可用',
+      'modelHint_f5': '未内置 — 离线神经语音运行时不可用',
       'dockLabel': '语音朗读',
       'dockPause': '暂停朗读',
       'dockResume': '继续朗读',
@@ -1100,7 +1169,7 @@ window.__ModuleLoader__.load({
               }, PROVIDERS.map((p) => React.createElement('option', { key: p, value: p }, p))),
               React.createElement('input', {
                 className: 'dts-model dts-input', value: row.model || '', disabled: !props.writable,
-                placeholder: MODEL_HINT[row.provider] || '', onChange: (e) => change(i, { model: e.target.value }),
+                placeholder: (typeof getModelHint === 'function' ? getModelHint(row.provider, t) : (MODEL_HINT[row.provider] || '')), onChange: (e) => change(i, { model: e.target.value }),
               }),
               React.createElement('input', {
                 className: 'dts-model dts-input', value: row.voice || '', disabled: !props.writable,
@@ -1668,6 +1737,7 @@ window.__ModuleLoader__.load({
     // looks like a working empty form.
       // Snapshot status contract (#163):
       // Without valid scope, or when loading/unavailable, editing is disabled.
+      const [revision, setRevision] = React.useState(null)
       const ready = !!scope && !!snap && snap.status === 'ready'
       const writable = ready && snap.writable !== false
 
@@ -1675,6 +1745,7 @@ window.__ModuleLoader__.load({
         const cfg = data && data.config ? data.config : {}
         setDraft(JSON.parse(JSON.stringify(cfg)))
         setCredentials(data && data.credentials ? data.credentials : {})
+        if (data && data.revision !== undefined) setRevision(data.revision)
         player.enabled = !!cfg.speakReplies
       }
 
@@ -1687,7 +1758,7 @@ window.__ModuleLoader__.load({
 
         fetch('/dsh-tts/integrations', { cache: 'no-store' }).then((res) => res.json()).then((data) => {
           if (!alive) return
-          if (data && data.ok) setIntegrations(data)
+          if (data && data.ok) setIntegrations(data.integrations || data)
         }).catch(() => {})
 
         const loadTelemetry = () => {
@@ -1768,10 +1839,14 @@ window.__ModuleLoader__.load({
             const v = String(keyDrafts[p] || '').trim()
             if (v) keys[p] = v
           })
+          const bodyPayload = Object.assign({}, draft, { keys: keys })
+          if (revision !== null && revision !== undefined) {
+            bodyPayload.revision = revision
+          }
           const res = await fetch('/dsh-tts/config', {
             method: 'PUT',
             headers: { 'Content-Type': 'application/json' },
-            body: JSON.stringify(Object.assign({}, draft, { keys: keys })),
+            body: JSON.stringify(bodyPayload),
           })
           const data = await res.json().catch(() => ({ /* parse error */ }))
           if (!res.ok) throw new Error((data && data.error && data.error.message) || ('HTTP ' + res.status))
@@ -1885,12 +1960,12 @@ window.__ModuleLoader__.load({
         React.createElement(VoiceDuplexEditor, {
           t: t, writable: writable, draft: draft,
           boolField: boolField, setTop: setTop,
-          installed: !!(integrations && integrations.voice && integrations.voice.installed),
+          installed: !!(integrations && (integrations.voice?.installed ?? integrations.integrations?.voice?.installed ?? integrations.voiceInstalled)),
         }),
         React.createElement(MessengerIntegrationEditor, {
           t: t, writable: writable, draft: draft,
           boolField: boolField, setTop: setTop,
-          installed: !!(integrations && integrations.messenger && integrations.messenger.installed),
+          installed: !!(integrations && (integrations.messenger?.installed ?? integrations.integrations?.messenger?.installed ?? integrations.messengerInstalled)),
         }),
         React.createElement('div', { className: 'dts-block' },
           React.createElement('div', { className: 'dts-h' }, t('pronTitle')),

package/lib/engines/manager.js

@@ -11,12 +11,12 @@ export const MODEL_SPECS = {
     files: [
       {
         name: 'kokoro-v0_19.onnx',
-        url: 'https://huggingface.co/hexgrad/Kokoro-82M/resolve/main/kokoro-v0_19.onnx',
+        url: 'https://huggingface.co/hexgrad/Kokoro-82M/resolve/f905c3175b9f7a77d4c1b9be894ff99ff7e9a8f2/kokoro-v0_19.onnx',
         minBytes: 1024,
       },
       {
         name: 'voices.json',
-        url: 'https://huggingface.co/hexgrad/Kokoro-82M/resolve/main/voices.json',
+        url: 'https://huggingface.co/hexgrad/Kokoro-82M/resolve/f905c3175b9f7a77d4c1b9be894ff99ff7e9a8f2/voices.json',
         minBytes: 1024,
       },
     ],
@@ -45,13 +45,26 @@ export function createModelManager({ root = process.env.DSH_HOME || path.join(os
     return path.join(baseDir, engineId)
   }
 
-  async function checkFileExists(filePath, minBytes = 1024) {
-    try {
-      const st = await fs.promises.stat(filePath)
-      return st.isFile() && st.size >= minBytes
-    } catch {
+  async function validateFileIntegrity(filePath, minBytes = 1024) {
+  try {
+    const st = await fs.promises.stat(filePath)
+    if (!st.isFile() || st.size < minBytes) return false
+    const fd = await fs.promises.open(filePath, 'r')
+    const buf = Buffer.alloc(Math.min(st.size, 1024))
+    await fd.read(buf, 0, buf.length, 0)
+    await fd.close()
+    const str = buf.toString('utf8').trim().toLowerCase()
+    if (str.startsWith('<!doctype') || str.startsWith('<html') || str.startsWith('<head') || str.includes('<html')) {
       return false
     }
+    return true
+  } catch (_err) {
+    return false
+  }
+}
+
+  async function checkFileExists(filePath, minBytes = 1024) {
+    return await validateFileIntegrity(filePath, minBytes)
   }
 
   async function getDirSize(dirPath) {
@@ -236,6 +249,10 @@ export function createModelManager({ root = process.env.DSH_HOME || path.join(os
           if (st.size < minExpected) {
             throw new Error(`File ${f.name} truncated: got ${st.size} bytes, expected at least ${minExpected}`)
           }
+          const valid = await validateFileIntegrity(tmpFile, minExpected)
+          if (!valid) {
+            throw new Error(`File ${f.name} failed integrity check: invalid format or HTML error`)
+          }
 
           await fs.promises.rename(tmpFile, targetFile)
           const idx = createdTmpFiles.indexOf(tmpFile)
@@ -255,7 +272,9 @@ export function createModelManager({ root = process.env.DSH_HOME || path.join(os
         lastErrors.set(engineId, { message: sanitized, failedAt: Date.now() })
         throw err
       } finally {
-        activeDownloads.delete(engineId)
+        if (activeDownloads.get(engineId) === downloadState) {
+          activeDownloads.delete(engineId)
+        }
       }
     }
 
@@ -269,7 +288,16 @@ export function createModelManager({ root = process.env.DSH_HOME || path.join(os
     const active = activeDownloads.get(engineId)
     if (active && active.controller) {
       active.controller.abort()
-      activeDownloads.delete(engineId)
+      if (active.promise) {
+        try {
+          await active.promise
+        } catch (_err) {
+          // aborted download rejected as expected
+        }
+      }
+      if (activeDownloads.get(engineId) === active) {
+        activeDownloads.delete(engineId)
+      }
       lastErrors.delete(engineId)
       return true
     }

package/lib/index.js

@@ -163,9 +163,9 @@ export const Config = z.object({
     .description('Longer replies are truncated before synthesis.'),
   chain: z.array(ChainEntry).volatile()
     .default([
-      { provider: 'edge', model: '', voice: 'ru-RU-SvetlanaNeural' },
+      { provider: 'edge', model: '', voice: 'en-US-JennyNeural' },
       { provider: 'piper', model: '', voice: '' },
-      { provider: 'espeak', model: '', voice: 'ru' },
+      { provider: 'espeak', model: '', voice: 'en' },
     ])
     .description('Fallback chain. Order is the order of attempts. A provider without a key is skipped.'),
   openaiKeyEnv: z.string().volatile().default('OPENAI_API_KEY'),
@@ -198,23 +198,56 @@ export const Config = z.object({
   maxQueue: z.number().step(1).min(0).volatile().default(8),
 })
 
+export function extractRevision(desc, targetNs = NS) {
+  if (!desc) return undefined
+  if (typeof desc.revision === 'number' || typeof desc.revision === 'string') {
+    return desc.revision
+  }
+  let list = null
+  if (Array.isArray(desc)) {
+    list = desc
+  } else if (Array.isArray(desc.namespaces)) {
+    list = desc.namespaces
+  } else if (Array.isArray(desc.entries)) {
+    list = desc.entries
+  }
+  if (list) {
+    const match = list.find((item) => {
+      if (!item || typeof item !== 'object') return false
+      const name = item.name || item.id || item.ns || item.namespace || item.key
+      return name === targetNs || name === 'goodandready/dsh-tts' || name === 'dsh-tts' || name === '@goodandready/dsh-tts'
+    })
+    if (match && (typeof match.revision === 'number' || typeof match.revision === 'string')) {
+      return match.revision
+    }
+    if (list.length === 1 && (typeof list[0].revision === 'number' || typeof list[0].revision === 'string')) {
+      return list[0].revision
+    }
+  }
+  if (typeof desc === 'object') {
+    for (const k of [targetNs, 'goodandready/dsh-tts', 'dsh-tts', '@goodandready/dsh-tts']) {
+      if (desc[k] && (typeof desc[k].revision === 'number' || typeof desc[k].revision === 'string')) {
+        return desc[k].revision
+      }
+    }
+  }
+  return undefined
+}
+
 let utteranceSeq = 0
 
 export function apply(ctx, config) {
   const providerBreaker = createProviderBreaker()
 
-  let savedConfig = null
   const live = () => {
     const fromBoxes = plainConfig(config)
-    const merged = savedConfig ? { ...fromBoxes, ...savedConfig } : fromBoxes
-    return plainConfig(Config(merged))
+    return plainConfig(Config(fromBoxes))
   }
 
   const settingsApi = {
     get: () => live(),
     replace: async (next, expectedRevision) => {
       const plain = plainConfig(next)
-      const prevSaved = savedConfig
       const svc = ctx?.get?.('settings') ?? ctx?.settings ?? ctx?.get?.('settingsForms') ?? ctx?.settingsForms
       if (!svc) {
         throw new Error('Settings service is unavailable')
@@ -223,28 +256,33 @@ export function apply(ctx, config) {
       if (rev === undefined && typeof svc.describe === 'function') {
         try {
           const desc = await svc.describe(NS)
-          if (desc && typeof desc.revision === 'number') {
-            rev = desc.revision
-          } else if (desc && Array.isArray(desc.namespaces)) {
-            const entry = desc.namespaces.find((n) => n.ns === NS)
-            if (entry && typeof entry.revision === 'number') rev = entry.revision
-          }
-        } catch {
+          rev = extractRevision(desc, NS)
+        } catch (_err) {
           // ignore describe error
         }
       }
-      try {
-        if (typeof svc.replace === 'function') {
-          await svc.replace(NS, plain, rev)
-        } else if (typeof svc.update === 'function') {
-          await svc.update(NS, plain, rev)
-        } else {
-          throw new Error('Settings service does not support replace or update')
+      if (typeof svc.replace === 'function') {
+        await svc.replace(NS, plain, rev)
+      } else if (typeof svc.update === 'function') {
+        await svc.update(NS, plain, rev)
+      } else {
+        throw new Error('Settings service does not support replace or update')
+      }
+
+      // Mutate volatile boxes directly so runtime dynamically reflects new settings (#152)
+      if (config && typeof config === 'object') {
+        for (const [k, v] of Object.entries(plain)) {
+          if (config[k] && typeof config[k] === 'object') {
+            const sym = Object.getOwnPropertySymbols(config[k]).find((s) => String(s).includes('volatile.write'))
+            if (sym && typeof config[k][sym] === 'function') {
+              try { config[k][sym](v) } catch (_wErr) { /* best effort */ }
+            } else if ('value' in config[k]) {
+              config[k].value = v
+            }
+          } else if (config[k] !== undefined) {
+            config[k] = v
+          }
         }
-        savedConfig = plain
-      } catch (err) {
-        savedConfig = prevSaved
-        throw err
       }
       return plain
     },
@@ -265,7 +303,7 @@ export function apply(ctx, config) {
     try {
       const resolved = await ctx.credentials.resolve(credentialRef(ref))
       if (resolved && resolved.value) return resolved.value
-    } catch { /* fall through to env */ }
+    } catch (_err) { /* fall through to env */ }
     return process.env[ref] || ''
   }
 
@@ -285,7 +323,7 @@ export function apply(ctx, config) {
       }
       const resolved = await ctx.credentials.resolve(credentialRef(ref))
       return { provider, ref, configured: !!(resolved && resolved.value), writable: true }
-    } catch {
+    } catch (_err) {
       return base
     }
   }
@@ -327,10 +365,21 @@ export function apply(ctx, config) {
   }
 
   async function configResponse() {
+    let revision = null
+    const svc = ctx?.get?.('settings') ?? ctx?.settings ?? ctx?.get?.('settingsForms') ?? ctx?.settingsForms
+    if (svc && typeof svc.describe === 'function') {
+      try {
+        const desc = await svc.describe(NS)
+        revision = extractRevision(desc, NS) ?? null
+      } catch (_err) {
+        // best effort describe
+      }
+    }
     return {
       ok: true,
       config: publicConfig(live()),
       credentials: await credentialsView(),
+      revision,
     }
   }
 
@@ -381,7 +430,7 @@ export function apply(ctx, config) {
           if (retold) {
             return speechPhrases(cfg.language).summaryIntro + ' ' + retold
           }
-        } catch { /* quiet fallback to truncation */ }
+        } catch (_err) { /* quiet fallback to truncation */ }
       }
       return text.slice(0, maxChars)
     }
@@ -389,7 +438,8 @@ export function apply(ctx, config) {
   }
 
   function synthesisFingerprint(text, provider, model, voice, role, ssmlStyle, cfg, lang) {
-    const parts = [text, provider, model || '', voice || '', role || '', ssmlStyle || '', lang || '']
+    const credRef = needsApiKey(provider) ? (keyEnvName(cfg, provider) || '') : ''
+    const parts = [text, provider, model || '', voice || '', role || '', ssmlStyle || '', lang || '', credRef]
     if (provider === 'custom') parts.push(cfg.customBaseUrl || '')
     else if (provider === 'openai') parts.push(cfg.openaiBaseUrl || '')
     else if (provider === 'mimo') parts.push(cfg.mimoBaseUrl || '', cfg.mimoFormat || '')
@@ -402,12 +452,13 @@ export function apply(ctx, config) {
 
   function flightFingerprint(text, order, models, voices, role, ssmlStyle, cfg, lang) {
     const details = order.map((p) => {
+      const credRef = needsApiKey(p) ? (keyEnvName(cfg, p) || '') : ''
       const extra = p === 'custom' ? (cfg.customBaseUrl || '') :
                     p === 'openai' ? (cfg.openaiBaseUrl || '') :
                     p === 'mimo' ? `${cfg.mimoBaseUrl || ''}:${cfg.mimoFormat || ''}` :
                     p === 'azure' ? (cfg.azureRegion || '') :
                     p === 'piper' ? `${cfg.piperModel || ''}:${cfg.piperBin || ''}` : ''
-      return `${p}:${models[p] || ''}:${voices[p] || ''}:${extra}`
+      return `${p}:${models[p] || ''}:${voices[p] || ''}:${credRef}:${extra}`
     }).join(';')
     return cacheKey([text, details, role, ssmlStyle, lang])
   }
@@ -656,7 +707,7 @@ export function apply(ctx, config) {
         if (p && p.id && activeReservations.has(p.id)) {
           const resv = activeReservations.get(p.id)
           resv.dropped = true
-          try { resv.controller.abort() } catch {}
+          try { resv.controller.abort() } catch (_abortErr) { /* ignore aborted controller */ }
           clearTimeout(resv.timer)
           activeReservations.delete(p.id)
         }
@@ -671,7 +722,7 @@ export function apply(ctx, config) {
         const resv = activeReservations.get(dropped.id)
         if (resv) {
           resv.dropped = true
-          try { resv.controller.abort() } catch {}
+          try { resv.controller.abort() } catch (_abortErr) { /* ignore aborted controller */ }
           clearTimeout(resv.timer)
           activeReservations.delete(dropped.id)
         }
@@ -720,8 +771,10 @@ export function apply(ctx, config) {
     if (!cfg.speakReplies) return
     const sid = session && session.id
     if (!sid) return
-    const isMessenger = session && (session.kind === 'im' || session.kind === 'messenger' || session.channel === 'messenger' || session.type === 'im')
-    if (isMessenger && cfg.messengerTtsEnabled === false) return
+    const isMessenger = session && (session.kind === 'im' || session.kind === 'messenger' || session.channel === 'messenger' || session.type === 'im' || session.source === 'messenger')
+    if (isMessenger && !cfg.messengerTtsEnabled) return
+    const isVoice = session && (session.kind === 'voice' || session.channel === 'voice' || session.type === 'voice' || session.mode === 'voice')
+    if (isVoice && !cfg.voiceDuplexEnabled) return
     // The agent is waiting for human approval — worth speaking aloud
     // so the user notices without watching the screen.
     if (event.type === 'approval/asked') {
@@ -737,7 +790,7 @@ export function apply(ctx, config) {
     }
 
     if (event.type === 'assistant/message') {
-      const text = assistantText(event.data && event.data.message)
+      const text = assistantText(event.data && (event.data.message ?? event.data.text ?? event.data))
       const agentName = (event.data && (
         event.data.subagent ||
         event.data.agent ||
@@ -750,8 +803,12 @@ export function apply(ctx, config) {
       // in sentence pieces; waiting for turn end means silence while the agent works.
       
       if (cfg.speakAsItGoes) {
-        const ready = cleanText(text, cfg, 0)
-        for (const piece of splitSentences(ready, cfg.sentenceChars)) speakPiece(sid, piece, cfg, activeRole)
+        prepareSpeechText(text, cfg).then((ready) => {
+          if (!ready) return
+          for (const piece of splitSentences(ready, cfg.sentenceChars)) speakPiece(sid, piece, cfg, activeRole)
+        }).catch((_err) => {
+          // best-effort speech fallback
+        })
         return
       }
       const cur = collectors.get(sid) || { parts: [], role: activeRole }
@@ -816,6 +873,7 @@ export function apply(ctx, config) {
           additionalProperties: false,
           properties: {
             ok: { type: 'boolean' },
+            id: { type: 'string' },
             provider: { type: 'string' },
             mime: { type: 'string' },
             bytes: { type: 'number' },

package/lib/providers/cloud.js

@@ -192,7 +192,15 @@ export function makeCloudProviders(deps, req, withCloudTimeout) {
     const m = mimeType.match(/rate=(\d+)/i)
     if (m) sampleRate = parseInt(m[1], 10)
     const isWavAlready = raw.length >= 4 && raw.toString('ascii', 0, 4) === 'RIFF'
-    const audio = isWavAlready ? raw : pcmToWav(raw, sampleRate, 1)
+    let audio = null
+    const lowerMime = mimeType.toLowerCase()
+    if (isWavAlready || lowerMime.includes('wav')) {
+      audio = raw
+    } else if (!lowerMime || lowerMime.includes('pcm') || lowerMime.includes('raw') || lowerMime.includes('l16')) {
+      audio = pcmToWav(raw, sampleRate, 1)
+    } else {
+      return { ok: false, provider: 'google', reason: `unsupported Google TTS MIME: ${mimeType}` }
+    }
     return { ok: audio.length > 0, provider: 'google', audio, mime: 'audio/wav', reason: audio.length ? '' : 'empty audio' }
     } finally {
       cleanup()

package/lib/providers.js

@@ -29,7 +29,7 @@ export function makeProviders(deps, req) {
       cleanup: () => {
         clearTimeout(timer)
         if (parentSignal && onParentAbort) {
-          try { parentSignal.removeEventListener('abort', onParentAbort) } catch {}
+          try { parentSignal.removeEventListener('abort', onParentAbort) } catch (_err) { /* parent abort listener already removed */ }
         }
       },
     }

package/lib/routes.js

@@ -119,7 +119,7 @@ export function registerHttpRoutes(ctx, api) {
         writeJson(res, 400, { ok: false, error: { code: 'param', message: 'missing engine parameter' } })
         return
       }
-      if (!MODEL_SPECS[engine]) {
+      if (!Object.hasOwn(MODEL_SPECS, engine)) {
         writeJson(res, 400, { ok: false, error: { code: 'param', message: 'unknown engine: ' + engine } })
         return
       }
@@ -160,7 +160,7 @@ export function registerHttpRoutes(ctx, api) {
         writeJson(res, 400, { ok: false, error: { code: 'param', message: 'missing engine parameter' } })
         return
       }
-      if (!MODEL_SPECS[engine]) {
+      if (!Object.hasOwn(MODEL_SPECS, engine)) {
         writeJson(res, 400, { ok: false, error: { code: 'param', message: 'unknown engine: ' + engine } })
         return
       }
@@ -187,7 +187,7 @@ export function registerHttpRoutes(ctx, api) {
         try {
           if (typeof ctx.get === 'function' && ctx.get(id)) return true
           if (ctx[id]) return true
-        } catch {}
+        } catch (_err) { /* service check fallback */ }
         try {
           const reg = ctx.registry || ctx.root?.registry
           if (reg && typeof reg.entries === 'function') {
@@ -197,7 +197,7 @@ export function registerHttpRoutes(ctx, api) {
               if (id === 'messenger' && (name.includes('messenger') || name.includes('dsh-im'))) return true
             }
           }
-        } catch {}
+        } catch (_err) { /* registry check fallback */ }
         try {
           if (ctx.webServer) {
             if (typeof ctx.webServer.hasRoute === 'function') {
@@ -207,7 +207,7 @@ export function registerHttpRoutes(ctx, api) {
               if (ctx.webServer.routes.some((r) => r && r.path && r.path.includes(id))) return true
             }
           }
-        } catch {}
+        } catch (_err) { /* webServer route check fallback */ }
         return false
       }
 
@@ -467,6 +467,11 @@ export function registerHttpRoutes(ctx, api) {
       const rawText = String(payload.text || '')
       if (!rawText.trim()) { writeJson(res, 400, { ok: false, error: { code: 'no-text', message: 'no text' } }); return }
       const cfg = live()
+      const isMessenger = payload.channel === 'messenger' || payload.source === 'messenger' || req.headers['x-dsh-channel'] === 'messenger'
+      if (isMessenger && !cfg.messengerTtsEnabled) {
+        writeJson(res, 403, { ok: false, error: { code: 'disabled', message: 'messenger tts is disabled' } })
+        return
+      }
       const controller = new AbortController()
       const timer = setTimeout(() => controller.abort(), cfg.timeoutMs)
       try {

package/lib/stream-hub.js

@@ -20,25 +20,22 @@ function detach(res) {
 }
 
 function safeWrite(res, payload) {
-  // If subscriber is already backpressured, check buffered limit
-  if (res.__dtsBackpressured) {
-    const currentBuffered = typeof res.writableLength === 'number'
-      ? res.writableLength
-      : (res.__dtsBufferedBytes || 0)
-    const payloadBytes = Buffer.byteLength(payload)
-    if (currentBuffered + payloadBytes > MAX_SUBSCRIBER_BUFFER) {
-      // Disconnect slow subscriber exceeding memory buffer limit (#186)
-      subscribers.delete(res)
-      detach(res)
-      try { res.end() } catch {}
-      return false
-    }
+  const payloadBytes = Buffer.byteLength(payload)
+  const currentBuffered = typeof res.writableLength === 'number'
+    ? res.writableLength
+    : (res.__dtsBufferedBytes || 0)
+
+  if (res.__dtsBackpressured && currentBuffered + payloadBytes > MAX_SUBSCRIBER_BUFFER) {
+    subscribers.delete(res)
+    detach(res)
+    try { res.end() } catch (_err) { /* already ended */ }
+    return false
   }
 
   let writeOk = true
   try {
     writeOk = res.write(payload)
-  } catch {
+  } catch (_err) {
     subscribers.delete(res)
     detach(res)
     return false
@@ -46,7 +43,13 @@ function safeWrite(res, payload) {
 
   if (writeOk === false) {
     res.__dtsBackpressured = true
-    res.__dtsBufferedBytes = (res.__dtsBufferedBytes || 0) + Buffer.byteLength(payload)
+    res.__dtsBufferedBytes = (res.__dtsBufferedBytes || 0) + payloadBytes
+    if (res.__dtsBufferedBytes > MAX_SUBSCRIBER_BUFFER) {
+      subscribers.delete(res)
+      detach(res)
+      try { res.end() } catch (_err) { /* already ended */ }
+      return false
+    }
     if (!res.__dtsOnDrain) {
       const onDrain = () => {
         res.__dtsBackpressured = false
@@ -55,7 +58,7 @@ function safeWrite(res, payload) {
       res.__dtsOnDrain = onDrain
       try {
         if (typeof res.on === 'function') res.on('drain', onDrain)
-      } catch { /* non-eventemitter mock */ }
+      } catch (_err) { /* non-eventemitter mock */ }
     }
   } else if (res.__dtsBackpressured && (!res.writableLength || res.writableLength === 0)) {
     res.__dtsBackpressured = false

package/lib/text.js

@@ -138,7 +138,13 @@ export function stripForSpeech(raw, maxChars, opts) {
 
 export function assistantText(message) {
   if (!message) return ''
-  return blocksToText(message.content)
+  if (typeof message === 'string') return message
+  if (typeof message === 'object') {
+    if (message.content !== undefined) return blocksToText(message.content)
+    if (typeof message.text === 'string') return message.text
+    if (typeof message.message === 'string') return message.message
+  }
+  return ''
 }
 
 // Protect common English abbreviations from false sentence splits

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-tts",
-  "version": "0.4.28",
+  "version": "0.4.29",
   "description": "Text-to-speech for DeepSeek Harness: speak agent replies in the Web UI with a provider fallback chain (OpenAI, ElevenLabs, Google, Azure, Groq, Deepgram, OpenRouter, Edge, Piper, eSpeak).",
   "license": "MIT",
   "type": "module",