dsh-dafeiyu-emotion 0.3.0 → 0.3.2

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # Changelog
 
+## 0.3.2
+
+- Added `ttsOnlyAfterVoice` (default `true`): spoken replies are limited to assistant messages within two minutes of a voice input, so normal coding chats do not read every answer aloud.
+- Added the matching settings toggle.
+
+## 0.3.1
+
+- Added spoken assistant replies. The host now forwards `assistant/message` text over `/plugins/dsh-dafeiyu-emotion/events`; the client plays it with TTS.
+- Added local Piper TTS (`zh_CN-huayan-medium`) as the default engine, with Edge TTS as an optional online engine and browser `speechSynthesis` as the final fallback.
+- Added `ttsEnabled`, `ttsEngine`, `ttsVoice`, `ttsRate`, `ttsPitch` and `ttsPiperModel` settings, including a voice selector and model path field.
+- Added `POST /plugins/dsh-dafeiyu-emotion/tts`, returning WAV for Piper and MP3 for Edge.
+- Edge TTS failures and missing Piper models fall back to local browser speech instead of breaking the reply.
+- Added TTS handler, assistant-text extraction, SSE broadcast and markdown-cleaning tests.
+
 ## 0.3.0
 
 - Added `empathyStyle` with four reply strategies: gentle, direct, humorous and concise. The choice is persisted and injected into the emotion-aware prompt.

package/README.md

@@ -76,6 +76,7 @@
 - 📈 多段语音的情绪时间线，可悬停查看每段文本、情绪和置信度；
 - 🔐 设置页内的隐私与运行状态：最近音频大小、模型加载、worker PID、临时文件清理；
 - 🔌 本机 `/context` 接口，其他插件可以读取最近一次情绪上下文；
+- 🔊 大肥鱼回复语音：Piper 本地模型默认朗读，Edge / 浏览器语音可选；
 - 🐋 大肥鱼主题的“鱼耳”入口，和普通对话共用同一个 composer；
 - 🧠 情绪标签：喜悦 / 低落 / 生气 / 紧张 / 惊讶 / 厌烦 / 平静；
 - 📈 额外给出 confidence、valence、arousal 和语气摘要；
@@ -115,13 +116,34 @@ dsh plugin --profile desktop add github:LiDu-china/dsh-dafeiyu-emotion
 ### 通过 Release tarball
 
 ```bash
-curl -L -o dsh-dafeiyu-emotion-0.3.0.tgz \
-  https://github.com/LiDu-china/dsh-dafeiyu-emotion/releases/download/v0.3.0/dsh-dafeiyu-emotion-0.3.0.tgz
-dsh plugin --profile desktop add "D:/path/to/dsh-dafeiyu-emotion-0.3.0.tgz"
+curl -L -o dsh-dafeiyu-emotion-0.3.2.tgz \
+  https://github.com/LiDu-china/dsh-dafeiyu-emotion/releases/download/v0.3.2/dsh-dafeiyu-emotion-0.3.2.tgz
+dsh plugin --profile desktop add "D:/path/to/dsh-dafeiyu-emotion-0.3.2.tgz"
 ```
 
 > DSH Desktop 的 profile 由 Electron 应用管理；命令行安装请使用桌面应用自带的 `dsh` 命令。
 
+## 大肥鱼语音回复
+
+助手回复生成后，插件会通过本机事件流把文本交给 TTS，让大肥鱼把回复读出来。
+
+默认使用 **Piper 本地模型**，音色是 `zh_CN-huayan-medium`。它不联网，延迟低，适合日常对话：
+
+```bash
+python -m pip install piper-tts
+```
+
+把模型放到默认目录：
+
+```text
+~/.dsh/dafeiyu-emotion/tts/piper/zh_CN-huayan-medium.onnx
+~/.dsh/dafeiyu-emotion/tts/piper/zh_CN-huayan-medium.onnx.json
+```
+
+也可以在设置页填写自定义 Piper 模型路径。模型缺失或合成失败时，会自动回退到浏览器 `speechSynthesis`，不会让回复静默丢失。
+
+如果更喜欢更自然的在线音色，可以把引擎切到 **Edge TTS**，默认音色 `zh-CN-XiaoyiNeural`。Edge 需要网络，且回复文本会发送到 Microsoft；如果网络或服务不可用，同样会回退到浏览器语音。
+
 ## 设置页与侧边栏
 
 插件会把自己的设置卡片注册到 DSH 的插件设置页。打开：
@@ -190,6 +212,13 @@ dsh plugin --profile desktop add "D:/path/to/dsh-dafeiyu-emotion-0.3.0.tgz"
 | `pythonPath` | 空 | 自定义 Python；空则自动寻找 `.venv-ser` |
 | `language` | `zh` | 识别语言：`zh` / `en` / `auto` |
 | `empathyStyle` | `gentle` | 共情策略：`gentle` / `direct` / `humorous` / `concise` |
+| `ttsEnabled` | `true` | 是否朗读大肥鱼回复 |
+| `ttsOnlyAfterVoice` | `true` | 只朗读录音后两分钟内的回复 |
+| `ttsEngine` | `piper` | `piper` 本地 / `edge` 在线 / `browser` 系统语音 |
+| `ttsVoice` | `zh-CN-XiaoyiNeural` | Edge TTS 音色 |
+| `ttsRate` | `+6%` | Piper/Edge 语速 |
+| `ttsPitch` | `+12Hz` | Edge 音调 |
+| `ttsPiperModel` | 空 | 自定义 Piper 模型路径；空则用默认目录 |
 | `maxAudioBytes` | 12 MB | 单次录音大小上限 |
 | `startupTimeoutMs` | 180000 | 首次加载增强后端的最长等待 |
 | `requestTimeoutMs` | 180000 | 单次分析请求超时 |

package/cordis.patch.yml

@@ -29,6 +29,15 @@
         pythonPath: ""
         language: zh
         empathyStyle: gentle
+        ttsEnabled: true
+        ttsOnlyAfterVoice: true
+        ttsEngine: piper
+        ttsVoice: zh-CN-XiaoyiNeural
+        ttsRate: "+6%"
+        ttsPitch: "+12Hz"
+        ttsCommand: ""
+        ttsPiperModel: ""
+        ttsPiperCommand: ""
         maxAudioBytes: 12582912
         startupTimeoutMs: 180000
         requestTimeoutMs: 180000

package/docs/MARKETPLACE.md

@@ -72,7 +72,7 @@ https://github.com/LiDu-china/dsh-dafeiyu-emotion/releases/download/v0.2.3/dsh-d
 dsh plugin --profile desktop add dsh-dafeiyu-emotion
 ```
 
-Published as `dsh-dafeiyu-emotion@0.3.0` on npm. The 1024Store entry refreshes its one-click install automatically. GitHub-source and Release-tarball installs remain available as fallbacks.
+Published as `dsh-dafeiyu-emotion@0.3.2` on npm. The 1024Store entry refreshes its one-click install automatically. GitHub-source and Release-tarball installs remain available as fallbacks.
 
 ## Privacy
 

package/docs/RELEASE-0.3.1.md

@@ -0,0 +1,39 @@
+## 0.3.1
+
+大肥鱼现在会说话了。
+
+### 语音回复
+
+- Host 监听 `assistant/message`，把回复文本通过 `/plugins/dsh-dafeiyu-emotion/events` 推给客户端。
+- 客户端收到后朗读回复，默认使用 **Piper 本地模型** `zh_CN-huayan-medium`。
+- 可选 **Edge TTS**（`zh-CN-XiaoyiNeural`，更自然，但需要网络），以及 **浏览器 speechSynthesis** 兜底。
+- Piper/Edge 失败时自动回退浏览器语音，不会让回复静默丢失。
+
+### 设置
+
+- 朗读开关、合成引擎、Edge 音色、语速、音调、Piper 模型路径。
+- 默认 `ttsEnabled: true`，Piper 模型不存在时自动使用浏览器语音。
+
+### 本地模型
+
+```bash
+python -m pip install piper-tts
+```
+
+默认模型目录：
+
+```text
+~/.dsh/dafeiyu-emotion/tts/piper/zh_CN-huayan-medium.onnx
+~/.dsh/dafeiyu-emotion/tts/piper/zh_CN-huayan-medium.onnx.json
+```
+
+可以在设置页填写自定义路径。
+
+### 接口
+
+```text
+POST /plugins/dsh-dafeiyu-emotion/tts
+GET  /plugins/dsh-dafeiyu-emotion/events
+```
+
+`/tts` 返回 WAV（Piper）或 MP3（Edge），只允许本机访问。

package/docs/RELEASE-0.3.2.md

@@ -0,0 +1,15 @@
+## 0.3.2
+
+语音回复的体验收口。
+
+- 新增 `ttsOnlyAfterVoice`，默认开启：只在录音后两分钟内的助手回复朗读。
+- 普通代码对话、没有录音触发的回复不再自动朗读。
+- 设置页新增「只朗读语音触发的回复」开关。
+
+## 安装
+
+```bash
+dsh plugin --profile desktop add dsh-dafeiyu-emotion
+```
+
+升级后请完全退出并重新启动 the agent harness。

package/lib/client.js

@@ -10,6 +10,8 @@ window.__ModuleLoader__.load({
     const ANALYZE_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/analyze'
     const CONFIG_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/config'
     const HEALTH_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/health'
+    const TTS_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/tts'
+    const EVENTS_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/events'
     const MAX_RECORDING_MS = 30000
     const PERMISSION_TIMEOUT_MS = 20000
     const EMOTION_LABELS = {
@@ -823,6 +825,13 @@ window.__ModuleLoader__.load({
         || draft.pythonPath !== config.pythonPath
         || draft.language !== config.language
         || draft.empathyStyle !== config.empathyStyle
+        || draft.ttsEnabled !== config.ttsEnabled
+        || draft.ttsOnlyAfterVoice !== config.ttsOnlyAfterVoice
+        || draft.ttsEngine !== config.ttsEngine
+        || draft.ttsVoice !== config.ttsVoice
+        || draft.ttsRate !== config.ttsRate
+        || draft.ttsPitch !== config.ttsPitch
+        || draft.ttsPiperModel !== config.ttsPiperModel
         || draft.maxAudioBytes !== config.maxAudioBytes
         || draft.startupTimeoutMs !== config.startupTimeoutMs
         || draft.requestTimeoutMs !== config.requestTimeoutMs
@@ -831,7 +840,7 @@ window.__ModuleLoader__.load({
       const save = async () => {
         if (!config || !draft || saving) return
         const patch = {}
-        for (const key of ['enabled', 'backendRoot', 'pythonPath', 'language', 'empathyStyle', 'maxAudioBytes', 'startupTimeoutMs', 'requestTimeoutMs']) {
+        for (const key of ['enabled', 'backendRoot', 'pythonPath', 'language', 'empathyStyle', 'ttsEnabled', 'ttsOnlyAfterVoice', 'ttsEngine', 'ttsVoice', 'ttsRate', 'ttsPitch', 'ttsPiperModel', 'maxAudioBytes', 'startupTimeoutMs', 'requestTimeoutMs']) {
           if (draft[key] !== config[key]) patch[key] = draft[key]
         }
         if (Object.keys(patch).length === 0) return
@@ -957,6 +966,7 @@ window.__ModuleLoader__.load({
         event.currentTarget.style.borderColor = T.border
       }
       const numberStyle = { ...inputStyle, width: 96, textAlign: 'right', fontVariantNumeric: 'tabular-nums' }
+      const selectStyle = { ...inputStyle, width: isSidebar ? '100%' : 220, paddingRight: 26 }
       const secondaryButton = {
         height: 32,
         padding: '0 11px',
@@ -994,6 +1004,15 @@ window.__ModuleLoader__.load({
           },
         }, value),
       )
+      const voiceOptions = [
+        ['zh-CN-XiaoyiNeural', '小艺 · 活泼'],
+        ['zh-CN-XiaoxiaoNeural', '晓晓 · 温柔'],
+        ['zh-CN-YunxiNeural', '云希 · 男声'],
+        ['zh-CN-liaoning-XiaobeiNeural', '小北 · 东北'],
+      ]
+      if (draft.ttsVoice && !voiceOptions.some(([value]) => value === draft.ttsVoice)) {
+        voiceOptions.push([draft.ttsVoice, draft.ttsVoice])
+      }
 
       return h(Root, {
         'data-testid': 'dsh-dafeiyu-emotion-settings',
@@ -1235,6 +1254,94 @@ window.__ModuleLoader__.load({
               onChange: (value) => update('empathyStyle', value),
             })),
           ),
+          h(SettingsSection, { title: '语音回复' },
+            h(SettingsField, {
+              icon: 'spark',
+              label: '朗读大肥鱼回复',
+              hint: '开启后，助手回复会自动朗读。Piper 本地模型最稳；Edge 需要网络；浏览器引擎留在本机。',
+            }, h(SettingsSwitch, {
+              checked: draft.ttsEnabled === true,
+              disabled,
+              label: '朗读大肥鱼回复',
+              onChange: (value) => update('ttsEnabled', value),
+            })),
+            h(SettingsField, {
+              icon: 'spark',
+              label: '只朗读语音触发的回复',
+              hint: '开启后只朗读录音后两分钟内的回复；关闭后所有助手回复都会朗读。',
+            }, h(SettingsSwitch, {
+              checked: draft.ttsOnlyAfterVoice !== false,
+              disabled: disabled || draft.ttsEnabled !== true,
+              label: '只朗读语音触发的回复',
+              onChange: (value) => update('ttsOnlyAfterVoice', value),
+            })),
+            h(SettingsField, {
+              icon: 'spark',
+              label: '合成引擎',
+              hint: 'Piper 使用本地中文模型；Edge 音色更自然但可能被网络限制；浏览器使用系统语音。',
+            }, h(SettingsSegmented, {
+              value: draft.ttsEngine || 'piper',
+              label: '合成引擎',
+              disabled,
+              options: [
+                { value: 'piper', label: 'Piper' },
+                { value: 'edge', label: 'Edge' },
+                { value: 'browser', label: '浏览器' },
+              ],
+              onChange: (value) => update('ttsEngine', value),
+            })),
+            h(SettingsField, {
+              icon: 'spark',
+              label: '音色',
+              hint: 'Edge TTS 的中文女声/男声；浏览器引擎使用系统默认音色。',
+              wide: true,
+            }, h('select', {
+              value: draft.ttsVoice || 'zh-CN-XiaoyiNeural',
+              disabled: disabled || draft.ttsEngine !== 'edge',
+              'aria-label': 'Edge TTS 音色',
+              onChange: (event) => update('ttsVoice', event.target.value),
+              style: selectStyle,
+            }, ...voiceOptions.map(([value, label]) => h('option', { key: value, value }, label)))),
+            draft.ttsEngine === 'piper' || draft.ttsEngine === 'edge'
+              ? h(SettingsField, {
+                  icon: 'folder',
+                  label: 'Piper 模型',
+                  hint: '留空使用 ~/.dsh/dafeiyu-emotion/tts/piper/zh_CN-huayan-medium.onnx。',
+                  wide: true,
+                }, h('input', {
+                  type: 'text',
+                  value: draft.ttsPiperModel || '',
+                  disabled: disabled || draft.ttsEngine !== 'piper',
+                  spellCheck: false,
+                  placeholder: '留空使用默认 Piper 模型',
+                  'aria-label': 'Piper 模型路径',
+                  onChange: (event) => update('ttsPiperModel', event.target.value),
+                  style: { ...inputStyle, width: '100%' },
+                }))
+              : null,
+            h(SettingsField, {
+              icon: 'spark',
+              label: '语速 / 音调',
+              hint: '语速 +6% 适用于 Piper/Edge；音调 +12Hz 仅 Edge。',
+            }, h('div', { style: { display: 'flex', alignItems: 'center', gap: 6, flexWrap: 'wrap' } },
+              h('input', {
+                type: 'text',
+                value: draft.ttsRate || '+6%',
+                disabled: disabled || draft.ttsEngine === 'browser',
+                'aria-label': '语速',
+                onChange: (event) => update('ttsRate', event.target.value),
+                style: { ...inputStyle, width: 76 },
+              }),
+              h('input', {
+                type: 'text',
+                value: draft.ttsPitch || '+12Hz',
+                disabled: disabled || draft.ttsEngine !== 'edge',
+                'aria-label': '音调',
+                onChange: (event) => update('ttsPitch', event.target.value),
+                style: { ...inputStyle, width: 86 },
+              }),
+            )),
+          ),
           h(SettingsSection, { title: '增强后端', hint: '可选' },
             h(SettingsField, {
               icon: 'folder',
@@ -1449,6 +1556,13 @@ window.__ModuleLoader__.load({
       const [sending, setSending] = useState(false)
       const [pluginEnabled, setPluginEnabled] = useState(null)
       const [language, setLanguage] = useState('zh')
+      const [ttsConfig, setTtsConfig] = useState({
+        enabled: false,
+        engine: 'piper',
+        voice: 'zh-CN-XiaoyiNeural',
+        rate: '+6%',
+        pitch: '+12Hz',
+      })
       const recorder = useRef(null)
       const abort = useRef(null)
       const timer = useRef(null)
@@ -1458,6 +1572,10 @@ window.__ModuleLoader__.load({
       const levelsRef = useRef([])
       const shortcutAction = useRef(null)
       const shortcutRegistration = useRef(null)
+      const ttsRef = useRef(ttsConfig)
+      const audioRef = useRef(null)
+      const speechGeneration = useRef(0)
+      ttsRef.current = ttsConfig
       const deliveryGate = useRef(null)
       if (!deliveryGate.current) deliveryGate.current = createDeliveryGate()
       const useInputSnapshot = typeof useInput === 'function' ? useInput : useNoInput
@@ -1734,12 +1852,95 @@ window.__ModuleLoader__.load({
         shortcutRegistration.current?.touch?.()
       }
 
+      function speakBrowser(text) {
+        const synth = window.speechSynthesis
+        if (!synth || typeof SpeechSynthesisUtterance === 'undefined') return false
+        synth.cancel()
+        const utterance = new SpeechSynthesisUtterance(text)
+        utterance.lang = 'zh-CN'
+        utterance.rate = 1.06
+        utterance.pitch = 1.08
+        synth.speak(utterance)
+        return true
+      }
+
+      async function speakReply(text) {
+        const config = ttsRef.current
+        if (!config?.enabled || !text) return false
+        const run = ++speechGeneration.current
+        if (config.engine === 'browser') return speakBrowser(text)
+        try {
+          const response = await fetch(TTS_ENDPOINT, {
+            method: 'POST',
+            headers: { 'content-type': 'application/json' },
+            body: JSON.stringify({
+              text,
+              voice: config.voice,
+              rate: config.rate,
+              pitch: config.pitch,
+            }),
+          })
+          if (!response.ok) {
+            const payload = await response.json().catch(() => ({}))
+            throw new Error(payload.error || `TTS 失败（${response.status}）`)
+          }
+          const blob = await response.blob()
+          if (run !== speechGeneration.current) return false
+          const url = URL.createObjectURL(blob)
+          audioRef.current?.pause?.()
+          const audio = new Audio(url)
+          audioRef.current = audio
+          audio.onended = () => {
+            URL.revokeObjectURL(url)
+            if (audioRef.current === audio) audioRef.current = null
+          }
+          try {
+            await audio.play()
+            return true
+          } catch {
+            URL.revokeObjectURL(url)
+            return speakBrowser(text)
+          }
+        } catch {
+          return speakBrowser(text)
+        }
+      }
+
+      useEffect(() => {
+        if (typeof EventSource === 'undefined') return undefined
+        const source = new EventSource(EVENTS_ENDPOINT)
+        source.onmessage = (event) => {
+          try {
+            const message = JSON.parse(event.data)
+            if (message?.kind !== 'assistant' || !message.text) return
+            if (message.sessionId && sessionId && message.sessionId !== sessionId) return
+            void speakReply(message.text)
+          } catch {}
+        }
+        return () => source.close()
+      }, [sessionId])
+
+      useEffect(() => () => {
+        speechGeneration.current += 1
+        audioRef.current?.pause?.()
+        window.speechSynthesis?.cancel?.()
+      }, [])
+
       useEffect(() => {
         let active = true
         const applyConfig = (next) => {
           if (!active || !next) return
           setPluginEnabled(next.enabled !== false)
           if (next.language) setLanguage(next.language)
+          setTtsConfig({
+            enabled: next.ttsEnabled === true,
+            engine: next.ttsEngine === 'browser'
+              ? 'browser'
+              : next.ttsEngine === 'piper' ? 'piper' : 'edge',
+            voice: next.ttsVoice || 'zh-CN-XiaoyiNeural',
+            rate: next.ttsRate || '+6%',
+            pitch: next.ttsPitch || '+12Hz',
+          })
         }
         fetch(CONFIG_ENDPOINT, { cache: 'no-store' })
           .then((response) => (response.ok ? response.json() : null))

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-dafeiyu-emotion",
-  "version": "0.3.0",
+  "version": "0.3.2",
   "description": "大肥鱼语音情绪耳朵：听语气、识别情绪，并让 DSH 按对方当下的状态自然回复。 · Voice + emotion-aware companion for DeepSeek Harness.",
   "type": "module",
   "main": "src/index.js",
@@ -61,7 +61,7 @@
   },
   "scripts": {
     "check": "node --check src/index.js && node --check lib/client.js",
-    "test": "node --check src/index.js && node --check lib/client.js && node --test scripts/test-config-schema.mjs scripts/test-config-handler.mjs scripts/test-analyze-handler.mjs scripts/test-backend-cancel.mjs scripts/test-delivery-policy.mjs scripts/test-shortcut-registration.mjs scripts/test-client-registration.mjs"
+    "test": "node --check src/index.js && node --check lib/client.js && node --test scripts/test-config-schema.mjs scripts/test-config-handler.mjs scripts/test-analyze-handler.mjs scripts/test-backend-cancel.mjs scripts/test-tts-handler.mjs scripts/test-delivery-policy.mjs scripts/test-shortcut-registration.mjs scripts/test-client-registration.mjs"
   },
   "dependencies": {
     "@deepseek-ai/schemastery": "3.18.5",

package/src/index.js

@@ -1,6 +1,6 @@
 import { spawn } from 'node:child_process'
 import { existsSync } from 'node:fs'
-import { mkdtemp, rm, writeFile } from 'node:fs/promises'
+import { mkdtemp, readFile, rm, writeFile } from 'node:fs/promises'
 import { createInterface } from 'node:readline'
 import { tmpdir } from 'node:os'
 import { dirname, join, resolve } from 'node:path'
@@ -14,6 +14,8 @@ const ANALYZE_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/analyze'
 const HEALTH_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/health'
 const CONFIG_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/config'
 const CONTEXT_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/context'
+const TTS_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/tts'
+const EVENTS_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/events'
 const SETTINGS_NAMESPACE = 'dsh-dafeiyu-emotion'
 
 const here = dirname(fileURLToPath(import.meta.url))
@@ -28,6 +30,15 @@ const DEFAULTS = Object.freeze({
   startupTimeoutMs: 180000,
   requestTimeoutMs: 180000,
   empathyStyle: 'gentle',
+  ttsEnabled: true,
+  ttsOnlyAfterVoice: true,
+  ttsEngine: 'piper',
+  ttsVoice: 'zh-CN-XiaoyiNeural',
+  ttsRate: '+6%',
+  ttsPitch: '+12Hz',
+  ttsCommand: '',
+  ttsPiperModel: '',
+  ttsPiperCommand: '',
 })
 
 const CONFIG_FIELDS = new Set(Object.keys(DEFAULTS))
@@ -69,6 +80,19 @@ export const Config = Schema.object({
     Schema.const('humorous').description('幽默'),
     Schema.const('concise').description('简洁'),
   ]).default('gentle').description('共情策略').volatile(),
+  ttsEnabled: Schema.boolean().default(true).description('朗读大肥鱼回复').volatile(),
+  ttsOnlyAfterVoice: Schema.boolean().default(true).description('只朗读语音触发的回复').volatile(),
+  ttsEngine: Schema.union([
+    Schema.const('piper').description('Piper 本地模型（推荐）'),
+    Schema.const('edge').description('Edge TTS（更好听，需要网络）'),
+    Schema.const('browser').description('浏览器本地语音'),
+  ]).default('piper').description('语音合成引擎').volatile(),
+  ttsVoice: Schema.string().default('zh-CN-XiaoyiNeural').description('Edge TTS 音色').volatile(),
+  ttsRate: Schema.string().default('+6%').description('语速').volatile(),
+  ttsPitch: Schema.string().default('+12Hz').description('音调').volatile(),
+  ttsCommand: Schema.string().default('').description('edge-tts 可执行文件路径').volatile(),
+  ttsPiperModel: Schema.string().default('').description('Piper 模型路径').volatile(),
+  ttsPiperCommand: Schema.string().default('').description('Piper 可执行文件路径').volatile(),
 }).description('大肥鱼语音情绪耳朵')
 
 const EMOTION_ZH = Object.freeze({
@@ -125,6 +149,266 @@ function jsonResponse(res, status, body) {
   res.end(payload)
 }
 
+function resolveEdgeTts(options) {
+  const explicit = String(options?.ttsCommand || '').trim()
+  if (explicit) return explicit
+  const fromEnv = String(process.env.EDGE_TTS_PATH || '').trim()
+  if (fromEnv) return fromEnv
+  const local = process.env.LOCALAPPDATA
+  const roaming = process.env.APPDATA
+  const candidates = [
+    local && join(local, 'Programs', 'Python', 'Python313', 'Scripts', 'edge-tts.exe'),
+    local && join(local, 'Programs', 'Python', 'Python312', 'Scripts', 'edge-tts.exe'),
+    roaming && join(roaming, 'Python', 'Python313', 'Scripts', 'edge-tts.exe'),
+    'edge-tts',
+  ].filter(Boolean)
+  return candidates.find((candidate) => candidate === 'edge-tts' || existsSync(candidate)) || 'edge-tts'
+}
+
+function defaultPiperModel() {
+  const home = process.env.USERPROFILE || process.env.HOME || ''
+  return join(home, '.dsh', 'dafeiyu-emotion', 'tts', 'piper', 'zh_CN-huayan-medium.onnx')
+}
+
+function resolvePiperModel(options) {
+  return String(options?.ttsPiperModel || process.env.DSH_DAFEIYU_PIPER_MODEL || defaultPiperModel()).trim()
+}
+
+function resolvePiperCommand(options) {
+  const explicit = String(options?.ttsPiperCommand || process.env.PIPER_PATH || '').trim()
+  if (explicit) return explicit
+  const local = process.env.LOCALAPPDATA
+  const candidates = [
+    local && join(local, 'Programs', 'Python', 'Python313', 'Scripts', 'piper.exe'),
+    local && join(local, 'Programs', 'Python', 'Python312', 'Scripts', 'piper.exe'),
+    'piper',
+  ].filter(Boolean)
+  return candidates.find((candidate) => candidate === 'piper' || existsSync(candidate)) || 'piper'
+}
+
+async function synthesizePiper(text, options) {
+  const model = resolvePiperModel(options)
+  const command = resolvePiperCommand(options)
+  if (!existsSync(model)) throw new Error(`Piper 模型不存在：${model}`)
+  const config = `${model}.json`
+  if (!existsSync(config)) throw new Error(`Piper 模型配置不存在：${config}`)
+  const dir = await mkdtemp(join(tmpdir(), 'dsh-dafeiyu-tts-'))
+  const input = join(dir, 'input.txt')
+  const out = join(dir, 'reply.wav')
+  try {
+    await writeFile(input, text, 'utf8')
+    const speed = Number.parseFloat(String(options.ttsRate || '+0%').replace('%', ''))
+    const lengthScale = Number.isFinite(speed) && speed !== 0 ? 1 / (1 + speed / 100) : 1
+    await new Promise((resolvePromise, rejectPromise) => {
+      const proc = spawn(command, [
+        '-m', model,
+        '-c', config,
+        '-i', input,
+        '-f', out,
+        '--length-scale', String(Math.max(0.7, Math.min(1.4, lengthScale))),
+      ], {
+        windowsHide: true,
+        env: { ...process.env, PYTHONUTF8: '1', PYTHONIOENCODING: 'utf-8' },
+      })
+      let stderr = ''
+      let settled = false
+      const timer = setTimeout(() => {
+        if (settled) return
+        settled = true
+        try { proc.kill() } catch {}
+        rejectPromise(new Error('Piper 合成超时'))
+      }, 30000)
+      proc.stderr?.on('data', (chunk) => { stderr += chunk.toString('utf8') })
+      proc.on('error', (error) => {
+        if (settled) return
+        settled = true
+        clearTimeout(timer)
+        rejectPromise(error)
+      })
+      proc.on('exit', (code) => {
+        if (settled) return
+        settled = true
+        clearTimeout(timer)
+        if (code === 0) resolvePromise()
+        else rejectPromise(new Error(stderr.trim() || `piper 退出（code=${code}）`))
+      })
+    })
+    return await readFile(out)
+  } finally {
+    await rm(dir, { recursive: true, force: true }).catch(() => {})
+  }
+}
+
+function ttsReady(options) {
+  if (!options.ttsEnabled) return false
+  if (options.ttsEngine === 'browser') return true
+  if (options.ttsEngine === 'piper') return existsSync(resolvePiperModel(options))
+  return existsSync(resolveEdgeTts(options))
+}
+
+function assistantTextFromEvent(event) {
+  const content = event?.data?.message?.content
+  if (!Array.isArray(content)) return ''
+  return content
+    .filter((block) => block?.type === 'text' && typeof block.text === 'string')
+    .map((block) => block.text)
+    .join('\n')
+}
+
+function cleanForSpeech(text) {
+  return String(text || '')
+    .replace(/```[\s\S]*?```/g, ' ')
+    .replace(/`[^`]*`/g, ' ')
+    .replace(/!\[[^\]]*\]\([^)]*\)/g, ' ')
+    .replace(/\[([^\]]+)\]\([^)]*\)/g, '$1')
+    .replace(/^#{1,6}\s+/gm, '')
+    .replace(/[*_~>#]/g, ' ')
+    .replace(/\s+/g, ' ')
+    .trim()
+    .slice(0, 800)
+}
+
+async function synthesizeEdge(text, options) {
+  const command = resolveEdgeTts(options)
+  const dir = await mkdtemp(join(tmpdir(), 'dsh-dafeiyu-tts-'))
+  const out = join(dir, 'reply.mp3')
+  try {
+    await new Promise((resolvePromise, rejectPromise) => {
+      const proc = spawn(command, [
+        '--voice', options.ttsVoice,
+        '--rate', options.ttsRate,
+        '--pitch', options.ttsPitch,
+        '--text', text,
+        '--write-media', out,
+      ], { windowsHide: true })
+      let stderr = ''
+      let settled = false
+      const timer = setTimeout(() => {
+        if (settled) return
+        settled = true
+        try { proc.kill() } catch {}
+        rejectPromise(new Error('TTS 合成超时'))
+      }, 25000)
+      proc.stderr?.on('data', (chunk) => { stderr += chunk.toString('utf8') })
+      proc.on('error', (error) => {
+        if (settled) return
+        settled = true
+        clearTimeout(timer)
+        rejectPromise(error)
+      })
+      proc.on('exit', (code) => {
+        if (settled) return
+        settled = true
+        clearTimeout(timer)
+        if (code === 0) resolvePromise()
+        else rejectPromise(new Error(stderr.trim() || `edge-tts 退出（code=${code}）`))
+      })
+    })
+    return await readFile(out)
+  } finally {
+    await rm(dir, { recursive: true, force: true }).catch(() => {})
+  }
+}
+
+function createTtsHandler(runtime) {
+  return async (req, res) => {
+    if (!isLocalRequest(req)) {
+      jsonResponse(res, 403, { error: '只允许本机访问' })
+      return
+    }
+    if (req.method !== 'POST') {
+      jsonResponse(res, 405, { error: 'method not allowed' })
+      return
+    }
+    if (!runtime.options.ttsEnabled) {
+      jsonResponse(res, 503, { error: '语音回复已关闭' })
+      return
+    }
+    try {
+      const body = await readJsonBody(req, 16 * 1024)
+      const text = cleanForSpeech(body?.text)
+      if (!text) throw new HttpError(400, '没有可朗读的文本')
+      if (runtime.options.ttsEngine === 'browser') {
+        throw new HttpError(400, '浏览器语音引擎在客户端本地运行，不需要调用 Host TTS')
+      }
+      const voice = String(body?.voice || runtime.options.ttsVoice)
+      const rate = String(body?.rate || runtime.options.ttsRate)
+      const pitch = String(body?.pitch || runtime.options.ttsPitch)
+      if (!/^[+-]\d{1,3}%$/.test(rate)) throw new HttpError(400, '语速格式不合法')
+      let audio
+      let contentType
+      if (runtime.options.ttsEngine === 'piper') {
+        audio = await synthesizePiper(text, { ...runtime.options, ttsRate: rate })
+        contentType = 'audio/wav'
+      } else {
+        if (!/^[A-Za-z]{2,3}-[A-Za-z0-9-]+Neural$/.test(voice)) throw new HttpError(400, '音色不合法')
+        if (!/^[+-]\d{1,3}Hz$/.test(pitch)) throw new HttpError(400, '音调格式不合法')
+        audio = await synthesizeEdge(text, {
+          ...runtime.options,
+          ttsVoice: voice,
+          ttsRate: rate,
+          ttsPitch: pitch,
+        })
+        contentType = 'audio/mpeg'
+      }
+      res.writeHead(200, {
+        'content-type': contentType,
+        'cache-control': 'no-store',
+        'content-length': audio.length,
+      })
+      res.end(audio)
+    } catch (error) {
+      const status = error instanceof HttpError ? error.status : 502
+      jsonResponse(res, status, {
+        error: error instanceof Error ? error.message : String(error),
+      })
+    }
+  }
+}
+
+function createSpeechStream() {
+  const clients = new Set()
+  return {
+    add(res) {
+      clients.add(res)
+      res.writeHead(200, {
+        'content-type': 'text/event-stream; charset=utf-8',
+        'cache-control': 'no-store',
+        connection: 'keep-alive',
+      })
+      res.write(': connected\n\n')
+      const timer = setInterval(() => {
+        try {
+          res.write(': ping\n\n')
+        } catch {
+          clearInterval(timer)
+          clients.delete(res)
+        }
+      }, 15000)
+      res.on('close', () => {
+        clearInterval(timer)
+        clients.delete(res)
+      })
+    },
+    broadcast(message) {
+      const payload = `data: ${JSON.stringify(message)}\n\n`
+      for (const res of clients) {
+        try {
+          res.write(payload)
+        } catch {
+          clients.delete(res)
+        }
+      }
+    },
+    close() {
+      for (const res of clients) {
+        try { res.end() } catch {}
+      }
+      clients.clear()
+    },
+  }
+}
+
 async function readJsonBody(req, limit) {
   const chunks = []
   let bytes = 0
@@ -181,6 +465,15 @@ function resolveOptions(raw = {}) {
     empathyStyle: ['gentle', 'direct', 'humorous', 'concise'].includes(value.empathyStyle)
       ? value.empathyStyle
       : DEFAULTS.empathyStyle,
+    ttsEnabled: value.ttsEnabled === true,
+    ttsOnlyAfterVoice: value.ttsOnlyAfterVoice !== false,
+    ttsEngine: ['piper', 'edge', 'browser'].includes(value.ttsEngine) ? value.ttsEngine : DEFAULTS.ttsEngine,
+    ttsVoice: String(value.ttsVoice || DEFAULTS.ttsVoice).trim(),
+    ttsRate: String(value.ttsRate || DEFAULTS.ttsRate).trim(),
+    ttsPitch: String(value.ttsPitch || DEFAULTS.ttsPitch).trim(),
+    ttsCommand: String(value.ttsCommand || '').trim(),
+    ttsPiperModel: String(value.ttsPiperModel || '').trim(),
+    ttsPiperCommand: String(value.ttsPiperCommand || '').trim(),
     maxAudioBytes: Math.min(32 * 1024 * 1024, Math.max(256 * 1024, asNumber(value.maxAudioBytes, DEFAULTS.maxAudioBytes))),
     startupTimeoutMs: Math.min(600000, Math.max(10000, asNumber(value.startupTimeoutMs, DEFAULTS.startupTimeoutMs))),
     requestTimeoutMs: Math.min(600000, Math.max(10000, asNumber(value.requestTimeoutMs, DEFAULTS.requestTimeoutMs))),
@@ -194,6 +487,15 @@ function publicConfig(options, writable = false) {
     pythonPath: options.pythonPath || '',
     language: options.language || DEFAULTS.language,
     empathyStyle: options.empathyStyle || DEFAULTS.empathyStyle,
+    ttsEnabled: options.ttsEnabled === true,
+    ttsOnlyAfterVoice: options.ttsOnlyAfterVoice !== false,
+    ttsEngine: options.ttsEngine || DEFAULTS.ttsEngine,
+    ttsVoice: options.ttsVoice || DEFAULTS.ttsVoice,
+    ttsRate: options.ttsRate || DEFAULTS.ttsRate,
+    ttsPitch: options.ttsPitch || DEFAULTS.ttsPitch,
+    ttsCommand: options.ttsCommand || '',
+    ttsPiperModel: options.ttsPiperModel || '',
+    ttsPiperCommand: options.ttsPiperCommand || '',
     maxAudioBytes: options.maxAudioBytes,
     startupTimeoutMs: options.startupTimeoutMs,
     requestTimeoutMs: options.requestTimeoutMs,
@@ -227,6 +529,59 @@ function normalizeConfigPatch(value) {
       patch.empathyStyle = style
       continue
     }
+    if (key === 'ttsEnabled') {
+      if (typeof next !== 'boolean') throw new HttpError(400, 'ttsEnabled 必须是布尔值')
+      patch.ttsEnabled = next
+      continue
+    }
+    if (key === 'ttsOnlyAfterVoice') {
+      if (typeof next !== 'boolean') throw new HttpError(400, 'ttsOnlyAfterVoice 必须是布尔值')
+      patch.ttsOnlyAfterVoice = next
+      continue
+    }
+    if (key === 'ttsEngine') {
+      const engine = String(next || '').trim().toLowerCase()
+      if (!['piper', 'edge', 'browser'].includes(engine)) {
+        throw new HttpError(400, 'ttsEngine 只能是 piper、edge 或 browser')
+      }
+      patch.ttsEngine = engine
+      continue
+    }
+    if (key === 'ttsVoice') {
+      const voice = String(next || '').trim()
+      if (!/^[A-Za-z]{2,3}-[A-Za-z0-9-]+Neural$/.test(voice)) {
+        throw new HttpError(400, 'ttsVoice 不是合法的 Edge TTS 音色')
+      }
+      patch.ttsVoice = voice
+      continue
+    }
+    if (key === 'ttsRate' || key === 'ttsPitch') {
+      const value = String(next || '').trim()
+      const pattern = key === 'ttsRate' ? /^[+-]\d{1,3}%$/ : /^[+-]\d{1,3}Hz$/
+      if (!pattern.test(value)) {
+        throw new HttpError(400, `${key} 格式应为 ${key === 'ttsRate' ? '+6%' : '+12Hz'}`)
+      }
+      patch[key] = value
+      continue
+    }
+    if (key === 'ttsCommand') {
+      if (typeof next !== 'string') throw new HttpError(400, 'ttsCommand 必须是字符串')
+      const text = next.trim()
+      if (text.length > 500 || /[\u0000-\u001f]/.test(text)) {
+        throw new HttpError(400, 'ttsCommand 不是合法的路径')
+      }
+      patch.ttsCommand = text
+      continue
+    }
+    if (key === 'ttsPiperModel' || key === 'ttsPiperCommand') {
+      if (typeof next !== 'string') throw new HttpError(400, `${key} 必须是字符串`)
+      const text = next.trim()
+      if (text.length > 500 || /[\u0000-\u001f]/.test(text)) {
+        throw new HttpError(400, `${key} 不是合法的路径`)
+      }
+      patch[key] = text
+      continue
+    }
     if (key === 'backendRoot' || key === 'pythonPath') {
       if (typeof next !== 'string') throw new HttpError(400, `${key} 必须是字符串`)
       const text = next.trim()
@@ -1012,6 +1367,7 @@ function createAnalyzeHandler(runtime, backend, getSpeechToText) {
       }
       runtime.lastAudioBytes = buffer.length
       runtime.lastAnalysisAt = runtime.lastAnalysis.at
+      runtime.lastVoiceAt = Date.now()
       jsonResponse(res, 200, analysis)
     } catch (error) {
       if (controller?.signal.aborted || res.writableEnded) return
@@ -1044,11 +1400,13 @@ export function apply(ctx, rawConfig = {}) {
     // shared runtime object and settings.update persists it for the next boot.
     options: resolveOptions(rawConfig),
     lastAnalysis: null,
+    lastVoiceAt: null,
     lastAudioBytes: 0,
     lastAnalysisAt: null,
     lastCleanupAt: null,
   }
   const backend = new VoiceBackend(runtime.options, logger)
+  const speechStream = createSpeechStream()
   let speechToText
   ctx.inject(['settings'], (settingsCtx) => {
     settingsService = settingsCtx.settings
@@ -1060,6 +1418,27 @@ export function apply(ctx, rawConfig = {}) {
     })
   })
   ctx.effect(() => () => backend.close())
+  ctx.effect(() => () => speechStream.close())
+
+  ctx.inject(['sessions'], (sessionCtx) => {
+    try {
+      const off = sessionCtx.on?.('session/event', (session, event) => {
+        if (event?.type !== 'assistant/message' || !runtime.options.ttsEnabled) return
+        if (runtime.options.ttsOnlyAfterVoice
+          && (!runtime.lastVoiceAt || Date.now() - runtime.lastVoiceAt > 120000)) return
+        const text = cleanForSpeech(assistantTextFromEvent(event))
+        if (!text) return
+        speechStream.broadcast({
+          kind: 'assistant',
+          sessionId: String(session?.header?.id ?? session?.id ?? ''),
+          text,
+        })
+      })
+      if (typeof off === 'function') ctx.effect(() => off)
+    } catch (error) {
+      logger.warn?.('[dsh-dafeiyu-emotion] failed to subscribe to session events:', error)
+    }
+  })
 
   ctx.inject(['webServer'], (httpCtx) => {
     httpCtx.effect(
@@ -1086,6 +1465,11 @@ export function apply(ctx, rawConfig = {}) {
             enabled: options.enabled,
             language: options.language,
             empathyStyle: options.empathyStyle,
+            ttsEnabled: options.ttsEnabled,
+            ttsOnlyAfterVoice: options.ttsOnlyAfterVoice,
+            ttsEngine: options.ttsEngine,
+            ttsVoice: options.ttsVoice,
+            ttsReady: ttsReady(options),
             engine: backendAvailable ? 'SenseVoice + emotion2vec' : 'DSH Speech + prosody',
             mode: !backendConfigured
               ? 'dsh-speech-prosody'
@@ -1101,6 +1485,7 @@ export function apply(ctx, rawConfig = {}) {
             privacy: {
               lastAudioBytes: runtime.lastAudioBytes,
               lastAnalysisAt: runtime.lastAnalysisAt,
+              lastVoiceAt: runtime.lastVoiceAt,
               lastCleanupAt: runtime.lastCleanupAt,
               workerPid: backendStatus.pid,
               modelLoadMs: backendStatus.modelLoadMs,
@@ -1154,19 +1539,52 @@ export function apply(ctx, rawConfig = {}) {
       }),
       'dsh-dafeiyu-emotion: 最近情绪上下文',
     )
+    httpCtx.effect(
+      () => httpCtx.webServer.register({
+        kind: 'exact',
+        path: TTS_ENDPOINT,
+        handler: createTtsHandler(runtime),
+      }),
+      'dsh-dafeiyu-emotion: 语音合成',
+    )
+    httpCtx.effect(
+      () => httpCtx.webServer.register({
+        kind: 'exact',
+        path: EVENTS_ENDPOINT,
+        handler(req, res) {
+          if (!isLocalRequest(req)) {
+            jsonResponse(res, 403, { error: '只允许本机访问' })
+            return
+          }
+          if (req.method !== 'GET') {
+            jsonResponse(res, 405, { error: 'method not allowed' })
+            return
+          }
+          speechStream.add(res)
+        },
+      }),
+      'dsh-dafeiyu-emotion: 回复语音事件',
+    )
   })
 
 }
 
 export {
   VoiceBackend,
+  assistantTextFromEvent,
   buildFallbackAnalysis,
   buildPrompt,
+  cleanForSpeech,
   createConfigHandler,
   decorateAnalysis,
   normalizeConfigPatch,
   plainConfig,
   publicConfig,
+  resolveEdgeTts,
+  resolvePiperCommand,
+  resolvePiperModel,
   resolveOptions,
+  synthesizePiper,
+  ttsReady,
   transcribeWithSpeech,
 }