dsh-dafeiyu-emotion 0.2.1 → 0.2.2

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # Changelog
 
+## 0.2.2
+
+- Fixed a duplicate-send race between auto-send and the manual send button.
+- Cancelled stale deliveries when the session changes or the component unmounts.
+- Stopped overwriting a busy composer when `insertText` is refused; send failures now keep the analysis for retry.
+- Added a cancel action and timeout while waiting for microphone permission.
+- Made prosody extraction yield to the UI thread instead of blocking on long recordings.
+- Corrected the base64 request-body limit and mapped validation failures to 400/413 instead of 500.
+- Pinned speech-provider readiness checks to the provider actually used for transcription.
+- Reported missing enhanced backends in health and preserved the real worker failure reason.
+- Added regression tests for delivery gating, shortcut availability and the analyze handler.
+
 ## 0.2.1
 
 - Added a native DSH voice shortcut, configurable from DSH keyboard settings.

package/README.md

@@ -90,6 +90,8 @@ dsh-dafeiyu-emotion
 
 `dsh-dafeiyu-emotion` 已发布到 npm。1024Store 会自动刷新一键安装入口；如果客户端尚未缓存到新版本，可以直接使用下面的 npm 命令。
 
+> `0.2.2` 目前通过 GitHub Release 提供；npm 上的最新版仍是 `0.2.1`，待 npm 发布完成后一键安装会自动升级。
+
 ### 通过 npm
 
 ```bash
@@ -107,9 +109,9 @@ dsh plugin --profile desktop add github:LiDu-china/dsh-dafeiyu-emotion
 ### 通过 Release tarball
 
 ```bash
-curl -L -o dsh-dafeiyu-emotion-0.2.1.tgz \
-  https://github.com/LiDu-china/dsh-dafeiyu-emotion/releases/download/v0.2.1/dsh-dafeiyu-emotion-0.2.1.tgz
-dsh plugin --profile desktop add "D:/path/to/dsh-dafeiyu-emotion-0.2.1.tgz"
+curl -L -o dsh-dafeiyu-emotion-0.2.2.tgz \
+  https://github.com/LiDu-china/dsh-dafeiyu-emotion/releases/download/v0.2.2/dsh-dafeiyu-emotion-0.2.2.tgz
+dsh plugin --profile desktop add "D:/path/to/dsh-dafeiyu-emotion-0.2.2.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。

package/backend/voice_worker.py

@@ -96,6 +96,7 @@ def main() -> int:
         if not line:
             continue
         request: dict[str, Any] = {}
+        request_id: Any = None
         try:
             request = json.loads(line)
             if request.get("type") == "shutdown":
@@ -175,15 +176,17 @@ def main() -> int:
             detail = traceback.format_exc()
             sys.stderr.write(detail)
             emit({
-                "id": request.get("id"),
+                "id": request_id,
                 "ok": False,
                 "error": f"{type(exc).__name__}: {exc}",
             })
 
     try:
         with contextlib.redirect_stdout(sys.stderr):
-            emotion.close()
-            asr.close()
+            try:
+                emotion.close()
+            finally:
+                asr.close()
     except Exception:
         sys.stderr.write(traceback.format_exc())
     return 0

package/docs/MARKETPLACE.md

@@ -68,7 +68,7 @@ https://raw.githubusercontent.com/LiDu-china/dsh-dafeiyu-emotion/main/assets/ico
 dsh plugin --profile desktop add dsh-dafeiyu-emotion
 ```
 
-Published as `dsh-dafeiyu-emotion@0.2.1` on npm. The 1024Store entry refreshes its one-click install automatically. GitHub-source install remains available as a fallback.
+The `0.2.2` build is attached to the GitHub Release. npm publication is pending, so npm and the 1024Store one-click entry currently serve `0.2.1`; they will refresh after the maintainer publishes `0.2.2`. GitHub-source and Release-tarball installs remain available as fallbacks.
 
 ## Privacy
 

package/lib/client.js

@@ -9,6 +9,7 @@ window.__ModuleLoader__.load({
 
     const ANALYZE_ENDPOINT = '/plugins/dsh-dafeiyu-emotion/analyze'
     const MAX_RECORDING_MS = 30000
+    const PERMISSION_TIMEOUT_MS = 20000
     const EMOTION_LABELS = {
       happy: '喜悦',
       sad: '低落',
@@ -29,6 +30,10 @@ window.__ModuleLoader__.load({
     }
     const voiceShortcutHandlers = []
 
+    function useNoInput() {
+      return null
+    }
+
     function registerVoiceShortcut(handler) {
       voiceShortcutHandlers.push(handler)
       return () => {
@@ -37,22 +42,52 @@ window.__ModuleLoader__.load({
       }
     }
 
-    function hasVoiceShortcutHandler() {
-      return voiceShortcutHandlers.length > 0
+    function canUseVoiceShortcut(phase, locked) {
+      return !locked && (phase === 'idle' || phase === 'recording' || phase === 'error')
+    }
+
+    function canInvokeVoiceShortcut() {
+      const handler = voiceShortcutHandlers.at(-1)
+      if (!handler) return false
+      try {
+        return handler.canHandle?.() === true
+      } catch {
+        return false
+      }
     }
 
     function invokeVoiceShortcut() {
       const handler = voiceShortcutHandlers.at(-1)
       if (!handler) return false
       try {
-        handler()
-        return true
+        return handler.run?.() === true
       } catch (error) {
         console.error('[dsh-dafeiyu-emotion] voice shortcut failed:', error)
         return false
       }
     }
 
+    function createDeliveryGate() {
+      let active = false
+      return {
+        begin() {
+          if (active) return false
+          active = true
+          return true
+        },
+        end() {
+          active = false
+        },
+        get active() {
+          return active
+        },
+      }
+    }
+
+    function isRunCurrent(run, generation) {
+      return run === generation.current
+    }
+
     const S = {
       trigger: {
         display: 'inline-flex',
@@ -214,7 +249,8 @@ window.__ModuleLoader__.load({
       return Math.sqrt(values.reduce((sum, value) => sum + (value - mean) ** 2, 0) / values.length)
     }
 
-    function computeProsody(samples, sampleRate) {
+    async function computeProsody(samples, sampleRate, shouldStop) {
+      if (shouldStop?.()) return {}
       const duration = samples.length / sampleRate
       if (samples.length < Math.round(sampleRate * 0.3)) {
         return { duration, voiced_ratio: 0, pause_ratio: 1, energy_n: 0.1, pitch_n: 0.5, var_n: 0.2 }
@@ -229,6 +265,10 @@ window.__ModuleLoader__.load({
       let frameIndex = 0
       let pitchFrames = 0
       for (let start = 0; start + frameLength <= samples.length; start += hop, frameIndex += 1) {
+        if ((frameIndex & 127) === 0) {
+          if (shouldStop?.()) return {}
+          await new Promise((resolve) => setTimeout(resolve, 0))
+        }
         let energy = 0
         let crossings = 0
         let previous = samples[start] >= 0
@@ -361,7 +401,7 @@ window.__ModuleLoader__.load({
         this.silentGain.connect(this.context.destination)
       }
 
-      async stop() {
+      async stop(isCancelled = () => false) {
         if (!this.active) return { wav: new Uint8Array(), prosody: {} }
         this.active = false
         const inputRate = this.sampleRate
@@ -374,10 +414,11 @@ window.__ModuleLoader__.load({
           offset += chunk.length
         }
         await this.dispose()
+        if (isCancelled()) return { wav: new Uint8Array(), prosody: {} }
         const resampled = resample(merged, inputRate, 16000)
         return {
           wav: encodeWav(resampled, 16000),
-          prosody: computeProsody(resampled, 16000),
+          prosody: await computeProsody(resampled, 16000, isCancelled),
         }
       }
 
@@ -420,7 +461,7 @@ window.__ModuleLoader__.load({
       )
     }
 
-    function VoiceFish({ sessionId, inputActions, locked, onActiveChange }) {
+    function VoiceFish({ sessionId, inputActions, useInput, locked, onActiveChange }) {
       const [phase, setPhase] = useState('idle')
       const [level, setLevel] = useState(0)
       const [levels, setLevels] = useState([])
@@ -428,13 +469,21 @@ window.__ModuleLoader__.load({
       const [message, setMessage] = useState('')
       const [analysis, setAnalysis] = useState(null)
       const [sent, setSent] = useState(false)
+      const [sending, setSending] = useState(false)
       const recorder = useRef(null)
       const abort = useRef(null)
       const timer = useRef(null)
       const ticker = useRef(null)
+      const permissionTimer = useRef(null)
       const generation = useRef(0)
       const levelsRef = useRef([])
       const shortcutAction = useRef(null)
+      const deliveryGate = useRef(null)
+      if (!deliveryGate.current) deliveryGate.current = createDeliveryGate()
+      const useInputSnapshot = typeof useInput === 'function' ? useInput : useNoInput
+      const input = useInputSnapshot((snapshot) => snapshot)
+      const inputRef = useRef(input)
+      inputRef.current = input
       const expanded = phase !== 'idle'
 
       useLayoutEffect(() => {
@@ -451,8 +500,10 @@ window.__ModuleLoader__.load({
       const clearTimers = () => {
         if (timer.current) clearTimeout(timer.current)
         if (ticker.current) clearInterval(ticker.current)
+        if (permissionTimer.current) clearTimeout(permissionTimer.current)
         timer.current = null
         ticker.current = null
+        permissionTimer.current = null
       }
 
       const reset = async () => {
@@ -471,6 +522,7 @@ window.__ModuleLoader__.load({
         setMessage('')
         setAnalysis(null)
         setSent(false)
+        setSending(false)
       }
 
       useEffect(() => {
@@ -498,14 +550,30 @@ window.__ModuleLoader__.load({
       const start = async () => {
         if (locked || recorder.current) return
         const run = ++generation.current
+        clearTimers()
         setMessage('')
         setAnalysis(null)
         setSent(false)
+        setSending(false)
         setPhase('requesting')
         const active = new Recorder()
         recorder.current = active
+        permissionTimer.current = setTimeout(() => {
+          permissionTimer.current = null
+          if (run !== generation.current) return
+          generation.current += 1
+          const pending = recorder.current
+          recorder.current = null
+          if (pending) pending.dispose().catch(() => {})
+          setPhase('error')
+          setMessage('麦克风授权超时，请检查浏览器权限后重试')
+        }, PERMISSION_TIMEOUT_MS)
         try {
           await active.start(onLevel)
+          if (permissionTimer.current) {
+            clearTimeout(permissionTimer.current)
+            permissionTimer.current = null
+          }
           if (run !== generation.current) {
             await active.dispose()
             return
@@ -520,8 +588,13 @@ window.__ModuleLoader__.load({
             void finish()
           }, MAX_RECORDING_MS)
         } catch (error) {
+          if (permissionTimer.current) {
+            clearTimeout(permissionTimer.current)
+            permissionTimer.current = null
+          }
           recorder.current = null
           await active.dispose().catch(() => {})
+          if (run !== generation.current) return
           setPhase('error')
           setMessage(error instanceof Error ? error.message : String(error))
         }
@@ -536,7 +609,7 @@ window.__ModuleLoader__.load({
         setPhase('analyzing')
         setMessage('大肥鱼正在听语气…')
         try {
-          const capture = await active.stop()
+          const capture = await active.stop(() => run !== generation.current)
           const wav = capture.wav
           if (run !== generation.current) return
           if (wav.length < 12000) {
@@ -562,7 +635,7 @@ window.__ModuleLoader__.load({
           setAnalysis(payload)
           setPhase('result')
           setMessage('')
-          await deliver(payload)
+          await deliver(payload, run)
         } catch (error) {
           if (error?.name === 'AbortError') return
           if (run !== generation.current) return
@@ -573,28 +646,58 @@ window.__ModuleLoader__.load({
         }
       }
 
-      const deliver = async (payload) => {
+      const deliver = async (payload, run = generation.current) => {
+        if (!isRunCurrent(run, generation)) return false
         const prompt = String(payload.prompt || '').trim()
         if (!prompt) {
-          setPhase('error')
-          setMessage('没有生成可发送的语音内容')
-          return
+          if (isRunCurrent(run, generation)) {
+            setPhase('error')
+            setMessage('没有生成可发送的语音内容')
+          }
+          return false
         }
+        const gate = deliveryGate.current
+        if (!gate.begin()) return false
+        setSending(true)
         try {
-          let inserted = false
-          try {
-            inserted = inputActions?.insertText?.(prompt, inputActions.captureInsertion()) === true
-          } catch {}
-          if (!inserted) inputActions?.setDraft?.(prompt)
+          if (!isRunCurrent(run, generation)) return false
+          const inputPhase = inputRef.current?.phase
+          if (inputPhase === 'adjudicating' || inputPhase === 'submitting') {
+            throw new Error('输入框正在提交，请稍后再试')
+          }
+          if (typeof inputActions?.insertText === 'function') {
+            if (typeof inputActions.captureInsertion !== 'function') {
+              throw new Error('当前 DSH 版本缺少输入位置接口')
+            }
+            const inserted = inputActions.insertText(prompt, inputActions.captureInsertion()) === true
+            if (!inserted) throw new Error('输入框正忙，请稍后再试')
+          } else if (typeof inputActions?.setDraft === 'function') {
+            inputActions.setDraft(prompt)
+          } else {
+            throw new Error('当前 DSH 版本不支持自动填入')
+          }
+          if (!isRunCurrent(run, generation)) return false
           await sleep(120)
-          inputActions?.submit?.()
+          if (!isRunCurrent(run, generation)) return false
+          if (typeof inputActions?.submit !== 'function') {
+            throw new Error('当前 DSH 版本不支持自动发送')
+          }
+          inputActions.submit()
+          if (!isRunCurrent(run, generation)) return false
           setSent(true)
-          setMessage(`听出${payload.emotionZh || '你的情绪'}，已交给大肥鱼回复`)
+          setMessage(`听出${payload.emotionZh || '你的情绪'}，已提交给大肥鱼`)
           await sleep(1350)
-          await reset()
+          if (isRunCurrent(run, generation)) await reset()
+          return true
         } catch (error) {
-          setPhase('error')
-          setMessage(`已识别，但发送失败：${error instanceof Error ? error.message : String(error)}`)
+          if (isRunCurrent(run, generation)) {
+            setPhase('error')
+            setMessage(`已识别，但发送失败：${error instanceof Error ? error.message : String(error)}`)
+          }
+          return false
+        } finally {
+          gate.end()
+          if (isRunCurrent(run, generation)) setSending(false)
         }
       }
 
@@ -606,17 +709,38 @@ window.__ModuleLoader__.load({
         void start()
       }
 
+      const retrySend = () => {
+        if (!analysis) return
+        setPhase('result')
+        setMessage('')
+        void deliver(analysis, generation.current)
+      }
+
       useEffect(() => {
-        shortcutAction.current = () => {
-          if (locked) return
-          if (phase === 'idle') void start()
-          else if (phase === 'recording') void finish()
-          else if (phase === 'error') retry()
+        shortcutAction.current = {
+          canHandle: () => canUseVoiceShortcut(phase, locked),
+          run: () => {
+            if (locked) return false
+            if (phase === 'idle') {
+              void start()
+              return true
+            }
+            if (phase === 'recording') {
+              void finish()
+              return true
+            }
+            if (phase === 'error') {
+              retry()
+              return true
+            }
+            return false
+          },
         }
       }, [locked, phase, sessionId])
 
-      useEffect(() => registerVoiceShortcut(() => {
-        shortcutAction.current?.()
+      useEffect(() => registerVoiceShortcut({
+        canHandle: () => shortcutAction.current?.canHandle?.() === true,
+        run: () => shortcutAction.current?.run?.() === true,
       }), [])
 
       if (phase === 'idle') {
@@ -675,13 +799,27 @@ window.__ModuleLoader__.load({
           'aria-label': '结束并分析',
           onClick: () => void finish(),
         }, '■'),
-        (phase === 'error' || phase === 'result') && h('button', {
+        phase === 'requesting' && h('button', {
+          type: 'button',
+          style: S.round,
+          title: '取消等待麦克风',
+          'aria-label': '取消等待麦克风',
+          onClick: () => void reset(),
+        }, '×'),
+        phase === 'error' && h('button', {
+          type: 'button',
+          style: S.round,
+          title: analysis ? '重试发送' : '重试录音',
+          'aria-label': analysis ? '重试发送' : '重试录音',
+          onClick: analysis ? retrySend : retry,
+        }, analysis ? '↺' : '↻'),
+        phase === 'result' && h('button', {
           type: 'button',
           style: S.round,
-          title: phase === 'error' ? '重试' : '关闭',
-          'aria-label': phase === 'error' ? '重试' : '关闭',
-          onClick: phase === 'error' ? retry : () => void reset(),
-        }, phase === 'error' ? '↻' : '×'),
+          title: '关闭',
+          'aria-label': '关闭',
+          onClick: () => void reset(),
+        }, '×'),
         phase === 'analyzing' && h('button', {
           type: 'button',
           style: S.round,
@@ -689,10 +827,11 @@ window.__ModuleLoader__.load({
           'aria-label': '取消分析',
           onClick: () => void reset(),
         }, '×'),
-        phase === 'result' && !sent && h('button', {
+        phase === 'result' && !sent && !sending && h('button', {
           type: 'button',
           style: { ...S.trigger, height: 26, fontSize: 12 },
-          onClick: () => void deliver(analysis),
+          disabled: sending,
+          onClick: () => void deliver(analysis, generation.current),
         }, '发送'),
       )
     }
@@ -712,7 +851,7 @@ window.__ModuleLoader__.load({
           },
           regions: ['page', 'editable'],
           modals: [],
-          resolve: () => hasVoiceShortcutHandler()
+          resolve: () => canInvokeVoiceShortcut()
             ? {
                 status: 'handled',
                 run: () => {
@@ -728,7 +867,7 @@ window.__ModuleLoader__.load({
             name: 'conversation.input.activity',
             id: 'dsh-dafeiyu-emotion',
             order: 35,
-            inject: () => ({}),
+            inject: (sessionId) => ({ sessionId }),
           }, VoiceFish)
         } catch (error) {
           console.error('[dsh-dafeiyu-emotion] failed to register voice activity:', error)
@@ -746,6 +885,11 @@ window.__ModuleLoader__.load({
       name: 'dsh-dafeiyu-emotion-client',
       inject: ['slots'],
       apply,
+      __test: {
+        createDeliveryGate,
+        canUseVoiceShortcut,
+        isRunCurrent,
+      },
     }
     return module.exports
   },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-dafeiyu-emotion",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "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",
@@ -53,17 +53,22 @@
     "url": "https://github.com/LiDu-china/dsh-dafeiyu-emotion/issues"
   },
   "license": "MIT",
+  "engines": {
+    "node": "^22.19.0 || >=24.0.0",
+    "dsh": ">=0.2.1-rc.2 <0.3.0"
+  },
   "publishConfig": {
     "registry": "https://registry.npmjs.org/"
   },
   "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-shortcut-registration.mjs"
+    "test": "node --check src/index.js && node --check lib/client.js && node --test scripts/test-shortcut-registration.mjs scripts/test-delivery-policy.mjs scripts/test-analyze-handler.mjs"
   },
   "dependencies": {
     "dsh-dafeiyu": "0.1.23"
   },
   "dsh": {
+    "manifestVersion": 1,
     "client": {
       "platform": "web",
       "inject": []

package/scripts/test-analyze-handler.mjs

@@ -0,0 +1,209 @@
+import assert from 'node:assert/strict'
+import { EventEmitter } from 'node:events'
+import { Readable } from 'node:stream'
+
+import { apply } from '../src/index.js'
+
+const ANALYZE = '/plugins/dsh-dafeiyu-emotion/analyze'
+const HEALTH = '/plugins/dsh-dafeiyu-emotion/health'
+
+function makeCtx(speechToText) {
+  const handlers = new Map()
+  const ctx = {
+    logger: { debug() {}, info() {}, error() {} },
+    effect(run) {
+      const dispose = run()
+      return typeof dispose === 'function' ? dispose : () => {}
+    },
+    inject(dependencies, callback) {
+      if (dependencies.includes('webServer')) {
+        callback({
+          effect(run) {
+            const dispose = run()
+            return typeof dispose === 'function' ? dispose : () => {}
+          },
+          webServer: {
+            register(spec) {
+              handlers.set(spec.path, spec.handler)
+              return () => {}
+            },
+          },
+        })
+      } else if (dependencies.includes('speechToText')) {
+        callback({
+          speechToText,
+          effect(run) {
+            const dispose = run()
+            return typeof dispose === 'function' ? dispose : () => {}
+          },
+        })
+      }
+    },
+  }
+  return { ctx, handlers }
+}
+
+function makeReq({ body, method = 'POST', remoteAddress = '127.0.0.1' } = {}) {
+  const payload = body === undefined ? '' : JSON.stringify(body)
+  const req = Readable.from([Buffer.from(payload)])
+  req.method = method
+  req.socket = { remoteAddress }
+  req.headers = { host: '127.0.0.1:13344' }
+  return req
+}
+
+function makeRes() {
+  const res = new EventEmitter()
+  res.writableEnded = false
+  res.status = 0
+  res.headers = null
+  res.payload = null
+  res.writeHead = (status, headers) => {
+    res.status = status
+    res.headers = headers
+  }
+  res.end = (payload) => {
+    res.payload = payload
+    res.writableEnded = true
+    res.emit('finish')
+  }
+  return res
+}
+
+function json(res) {
+  return JSON.parse(res.payload)
+}
+
+function validWav() {
+  const wav = Buffer.alloc(500)
+  wav.write('RIFF', 0, 'ascii')
+  wav.write('WAVE', 8, 'ascii')
+  return wav
+}
+
+async function call(handler, req) {
+  const res = makeRes()
+  await handler(req, res)
+  return res
+}
+
+// Non-local requests are refused before the body is read.
+{
+  const { ctx, handlers } = makeCtx(undefined)
+  apply(ctx, { enabled: true, backendRoot: '' })
+  const res = await call(handlers.get(ANALYZE), makeReq({
+    body: { audioBase64: 'AA==' },
+    remoteAddress: '10.0.0.5',
+  }))
+  assert.equal(res.status, 403)
+  assert.equal(json(res).error, '只允许本机访问')
+}
+
+// Wrong method is rejected.
+{
+  const { ctx, handlers } = makeCtx(undefined)
+  apply(ctx, { enabled: true, backendRoot: '' })
+  const res = await call(handlers.get(ANALYZE), makeReq({ method: 'GET' }))
+  assert.equal(res.status, 405)
+}
+
+// Body too large is 413, not 500.
+{
+  const { ctx, handlers } = makeCtx(undefined)
+  apply(ctx, { enabled: true, backendRoot: '', maxAudioBytes: 256 * 1024 })
+  const res = await call(handlers.get(ANALYZE), makeReq({
+    body: { audioBase64: 'A'.repeat(500000) },
+  }))
+  assert.equal(res.status, 413)
+  assert.match(json(res).error, /请求体过大/)
+}
+
+// Invalid JSON and invalid WAV are 400.
+{
+  const { ctx, handlers } = makeCtx(undefined)
+  apply(ctx, { enabled: true, backendRoot: '' })
+  const invalidJson = await call(handlers.get(ANALYZE), makeReq({ body: undefined }))
+  assert.equal(invalidJson.status, 400)
+  const invalidWav = await call(handlers.get(ANALYZE), makeReq({
+    body: { audioBase64: Buffer.alloc(500).toString('base64') },
+  }))
+  assert.equal(invalidWav.status, 400)
+  assert.match(json(invalidWav).error, /WAV/)
+}
+
+// The readiness check and resolve must use the selected provider.
+{
+  let resolveArgs
+  const speechToText = {
+    snapshot: () => ({
+      providers: [
+        { id: 'first', preparation: { phase: 'ready' } },
+        { id: 'second', preparation: { phase: 'ready' } },
+      ],
+      selection: { providerId: 'second' },
+    }),
+    resolve(args) {
+      resolveArgs = args
+      return { provider: { id: args.providerId }, audio: args.audio, language: args.language }
+    },
+    async transcribe() {
+      return { text: '你好' }
+    },
+  }
+  const { ctx, handlers } = makeCtx(speechToText)
+  apply(ctx, { enabled: true, backendRoot: '' })
+  const res = await call(handlers.get(ANALYZE), makeReq({
+    body: { audioBase64: validWav().toString('base64'), prosody: {} },
+  }))
+  assert.equal(res.status, 200)
+  assert.equal(resolveArgs.providerId, 'second')
+  assert.equal(json(res).transcript, '你好')
+}
+
+// A configured but missing backend is visible in both analyze and health.
+{
+  const missing = 'Z:/dsh-dafeiyu-emotion-missing-backend'
+  const speechToText = {
+    snapshot: () => ({ providers: [], selection: {} }),
+    resolve() {
+      throw new Error('should not resolve')
+    },
+    async transcribe() {
+      throw new Error('speech unavailable')
+    },
+  }
+  const { ctx, handlers } = makeCtx(speechToText)
+  apply(ctx, { enabled: true, backendRoot: missing })
+  const analyze = await call(handlers.get(ANALYZE), makeReq({
+    body: { audioBase64: validWav().toString('base64'), prosody: {} },
+  }))
+  assert.equal(analyze.status, 500)
+  assert.match(json(analyze).error, /语音后端目录不存在/)
+
+  const health = await call(handlers.get(HEALTH), makeReq({ method: 'GET' }))
+  const payload = json(health)
+  assert.equal(payload.mode, 'enhanced-backend-missing')
+  assert.equal(payload.backendConfigured, true)
+  assert.equal(payload.backendAvailable, false)
+  assert.match(payload.lastError, /语音后端目录不存在/)
+}
+
+// Health reflects real provider readiness, not just service presence.
+{
+  const speechToText = {
+    snapshot: () => ({
+      providers: [{ id: 'sensevoice', preparation: { phase: 'ready' } }],
+      selection: { providerId: 'sensevoice' },
+    }),
+  }
+  const { ctx, handlers } = makeCtx(speechToText)
+  apply(ctx, { enabled: true, backendRoot: '' })
+  const health = await call(handlers.get(HEALTH), makeReq({ method: 'GET' }))
+  const payload = json(health)
+  assert.equal(payload.mode, 'dsh-speech-prosody')
+  assert.equal(payload.dshSpeechReady, true)
+  assert.equal(payload.speechProvider, 'sensevoice')
+  assert.equal(payload.speechPhase, 'ready')
+}
+
+console.log('analyze handler ok')

package/scripts/test-delivery-policy.mjs

@@ -0,0 +1,47 @@
+import assert from 'node:assert/strict'
+
+let clientModule
+
+globalThis.window = {
+  __ModuleLoader__: {
+    load({ factory }) {
+      clientModule = factory((id) => {
+        assert.equal(id, 'react')
+        return {
+          createElement() {},
+          Fragment: {},
+          useEffect() {},
+          useLayoutEffect() {},
+          useRef() {},
+          useState() {},
+        }
+      })
+    },
+  },
+}
+
+await import(new URL('../lib/client.js', import.meta.url).href)
+assert.ok(clientModule?.__test, 'client bundle should expose delivery test helpers')
+
+const { createDeliveryGate, canUseVoiceShortcut, isRunCurrent } = clientModule.__test
+
+const gate = createDeliveryGate()
+assert.equal(gate.begin(), true, 'first delivery should acquire the gate')
+assert.equal(gate.begin(), false, 'a concurrent delivery must be rejected')
+gate.end()
+assert.equal(gate.begin(), true, 'the gate should reopen after settlement')
+gate.end()
+
+assert.equal(canUseVoiceShortcut('idle', false), true)
+assert.equal(canUseVoiceShortcut('recording', false), true)
+assert.equal(canUseVoiceShortcut('error', false), true)
+assert.equal(canUseVoiceShortcut('requesting', false), false)
+assert.equal(canUseVoiceShortcut('analyzing', false), false)
+assert.equal(canUseVoiceShortcut('result', false), false)
+assert.equal(canUseVoiceShortcut('idle', true), false)
+
+assert.equal(isRunCurrent(2, { current: 2 }), true)
+assert.equal(isRunCurrent(2, { current: 3 }), false)
+
+delete globalThis.window
+console.log('delivery policy ok')

package/scripts/test-shortcut-registration.mjs

@@ -60,6 +60,9 @@ assert.deepEqual(registeredCommand?.defaults['web:windows'], {
   modifiers: ['primary', 'alt'],
 })
 assert.deepEqual(registeredCommand?.resolve(), { status: 'pass' })
+assert.equal(clientModule.__test.canUseVoiceShortcut('idle', false), true)
+assert.equal(clientModule.__test.canUseVoiceShortcut('analyzing', false), false)
+assert.equal(clientModule.__test.canUseVoiceShortcut('recording', true), false)
 
 delete globalThis.window
 console.log('shortcut registration ok')

package/src/index.js

@@ -35,6 +35,20 @@ const EMOTION_ZH = Object.freeze({
   neutral: '平静',
 })
 
+class HttpError extends Error {
+  constructor(status, message) {
+    super(message)
+    this.name = 'HttpError'
+    this.status = status
+  }
+}
+
+function createAbortError() {
+  const error = new Error('已取消')
+  error.name = 'AbortError'
+  return error
+}
+
 function isLocalRequest(req) {
   if (!['127.0.0.1', '::1', '::ffff:127.0.0.1'].includes(req.socket?.remoteAddress)) return false
   if (req.headers?.host) {
@@ -55,6 +69,7 @@ function isLocalRequest(req) {
 }
 
 function jsonResponse(res, status, body) {
+  if (res.writableEnded || res.destroyed) return
   const payload = JSON.stringify(body)
   res.writeHead(status, {
     'content-type': 'application/json; charset=utf-8',
@@ -69,10 +84,14 @@ async function readJsonBody(req, limit) {
   let bytes = 0
   for await (const chunk of req) {
     bytes += chunk.length
-    if (bytes > limit) throw new Error('请求体过大，请缩短录音后重试')
+    if (bytes > limit) throw new HttpError(413, '请求体过大，请缩短录音后重试')
     chunks.push(chunk)
   }
-  return JSON.parse(Buffer.concat(chunks).toString('utf8'))
+  try {
+    return JSON.parse(Buffer.concat(chunks).toString('utf8'))
+  } catch {
+    throw new HttpError(400, '请求体不是合法的 JSON')
+  }
 }
 
 function asNumber(value, fallback) {
@@ -325,6 +344,7 @@ class VoiceBackend {
     this.proc = null
     this.ready = null
     this.startup = null
+    this.startupAbort = null
     this.pending = new Map()
     this.nextId = 0
     this.queue = Promise.resolve()
@@ -332,6 +352,39 @@ class VoiceBackend {
     this.lastError = null
   }
 
+  noteError(message) {
+    this.lastError = message
+  }
+
+  /**
+   * Stop one worker deliberately. Unlike an unexpected exit this keeps the
+   * existing lastError (a timeout/cancel reason) instead of overwriting it
+   * with a generic "worker exited" message.
+   */
+  stopProcess(proc, { error, preserveLastError = false } = {}) {
+    if (this.proc !== proc) return
+    this.startupAbort?.()
+    this.proc = null
+    this.ready = false
+    if (!preserveLastError && error) this.lastError = error.message
+    const pending = [...this.pending.entries()]
+    this.pending.clear()
+    for (const [, slot] of pending) {
+      clearTimeout(slot.timer)
+      slot.reject(error || new Error('语音后端已停止'))
+    }
+    try {
+      proc.kill()
+    } catch {}
+  }
+
+  /** Abort the in-flight worker so a cancelled request cannot block the queue. */
+  cancel() {
+    const proc = this.proc
+    if (!proc) return
+    this.stopProcess(proc, { error: createAbortError(), preserveLastError: true })
+  }
+
   status() {
     return {
       state: this.proc && !this.proc.killed ? (this.ready ? 'ready' : 'starting') : 'sleeping',
@@ -347,11 +400,13 @@ class VoiceBackend {
     if (this.startup) return this.startup
     this.startup = new Promise((resolvePromise, rejectPromise) => {
       if (!existsSync(this.options.backendRoot)) {
-        rejectPromise(new Error(`语音后端目录不存在：${this.options.backendRoot}`))
+        this.lastError = `语音后端目录不存在：${this.options.backendRoot}`
+        rejectPromise(new Error(this.lastError))
         return
       }
       if (!existsSync(defaultWorker)) {
-        rejectPromise(new Error(`插件自带 worker 缺失：${defaultWorker}`))
+        this.lastError = `插件自带 worker 缺失：${defaultWorker}`
+        rejectPromise(new Error(this.lastError))
         return
       }
       const proc = spawn(this.python, [defaultWorker, this.options.backendRoot], {
@@ -374,18 +429,21 @@ class VoiceBackend {
       const timer = setTimeout(() => {
         if (settled) return
         settled = true
+        const error = new Error(`语音模型加载超时（>${Math.round(this.options.startupTimeoutMs / 1000)} 秒）`)
         this.lastError = '语音模型加载超时'
-        proc.kill()
-        rejectPromise(new Error(`${this.lastError}（>${Math.round(this.options.startupTimeoutMs / 1000)} 秒）`))
+        this.stopProcess(proc, { error, preserveLastError: true })
+        rejectPromise(error)
       }, this.options.startupTimeoutMs)
 
       const finishReady = (error) => {
+        if (this.proc !== proc) return
         if (settled) return
         settled = true
+        this.startupAbort = null
         clearTimeout(timer)
         if (error) {
           this.lastError = error.message
-          proc.kill()
+          this.stopProcess(proc, { error, preserveLastError: true })
           rejectPromise(error)
           return
         }
@@ -394,6 +452,14 @@ class VoiceBackend {
         resolvePromise()
       }
 
+      this.startupAbort = () => {
+        if (settled) return
+        settled = true
+        this.startupAbort = null
+        clearTimeout(timer)
+        rejectPromise(createAbortError())
+      }
+
       const lines = createInterface({ input: proc.stdout })
       lines.on('line', (line) => {
         const text = line.trim()
@@ -435,28 +501,35 @@ class VoiceBackend {
       })
 
       const fail = (error) => {
-        this.ready = false
-        this.lastError = error.message
-        if (!settled) finishReady(error)
-        for (const slot of this.pending.values()) {
-          clearTimeout(slot.timer)
-          slot.reject(error)
+        if (this.proc !== proc) return
+        if (!settled) {
+          finishReady(error)
+          return
         }
-        this.pending.clear()
-        this.proc = null
-        this.startup = null
+        this.lastError = error.message
+        this.stopProcess(proc, { error, preserveLastError: true })
       }
       proc.on('error', fail)
       proc.on('exit', (code) => {
-        if (!settled) finishReady(new Error(`语音后端退出（code=${code}）`))
-        else if (this.proc === proc) {
-          const detail = this.stderr.slice(-6).join('\n')
-          fail(new Error(`语音后端意外退出（code=${code}）${detail ? `\n${detail}` : ''}`))
+        if (this.proc !== proc) return
+        if (!settled) {
+          finishReady(new Error(`语音后端退出（code=${code}）`))
+          return
         }
+        const detail = this.stderr.slice(-6).join('\n')
+        const error = new Error(`语音后端意外退出（code=${code}）${detail ? `\n${detail}` : ''}`)
+        this.lastError = error.message
+        this.stopProcess(proc, { error, preserveLastError: true })
+      })
+      proc.stdin.on('error', (error) => {
+        if (this.proc !== proc) return
+        const wrapped = error instanceof Error ? error : new Error(String(error))
+        this.lastError = `语音后端写入失败：${wrapped.message}`
+        this.stopProcess(proc, { error: wrapped, preserveLastError: true })
       })
-      proc.stdin.on('error', () => {})
     }).finally(() => {
       this.startup = null
+      this.startupAbort = null
     })
     return this.startup
   }
@@ -467,16 +540,36 @@ class VoiceBackend {
       const proc = this.proc
       if (!proc || proc.killed || !proc.stdin) throw new Error('语音后端不可用')
       const id = ++this.nextId
+      let settled = false
+      let timer
       const result = new Promise((resolvePromise, rejectPromise) => {
-        const timer = setTimeout(() => {
+        const finish = (callback, value) => {
+          if (settled) return
+          settled = true
+          if (timer) clearTimeout(timer)
           this.pending.delete(id)
+          callback(value)
+        }
+        timer = setTimeout(() => {
+          const error = new Error(`语音分析超时（>${Math.round(this.options.requestTimeoutMs / 1000)} 秒）`)
           this.lastError = '语音分析超时'
-          proc.kill()
-          rejectPromise(new Error(`${this.lastError}（>${Math.round(this.options.requestTimeoutMs / 1000)} 秒）`))
+          this.stopProcess(proc, { error, preserveLastError: true })
         }, this.options.requestTimeoutMs)
-        this.pending.set(id, { resolve: resolvePromise, reject: rejectPromise, timer })
+        this.pending.set(id, {
+          resolve: (message) => finish(resolvePromise, message),
+          reject: (error) => finish(rejectPromise, error),
+          timer,
+        })
       })
-      proc.stdin.write(`${JSON.stringify({ id, wav: wavPath, language: this.options.language })}\n`)
+      try {
+        proc.stdin.write(`${JSON.stringify({ id, wav: wavPath, language: this.options.language })}\n`)
+      } catch (error) {
+        const wrapped = error instanceof Error ? error : new Error(String(error))
+        this.lastError = `语音后端写入失败：${wrapped.message}`
+        this.stopProcess(proc, { error: wrapped, preserveLastError: true })
+        result.catch(() => {})
+        throw wrapped
+      }
       return result
     }
     const task = this.queue.then(run, run)
@@ -490,6 +583,7 @@ class VoiceBackend {
 
   async close() {
     const proc = this.proc
+    this.startupAbort?.()
     this.proc = null
     this.ready = false
     this.startup = null
@@ -515,7 +609,8 @@ function speechProviderReady(provider) {
   return phase === undefined || ['ready', 'standby', 'waking'].includes(phase)
 }
 
-async function transcribeWithSpeech(audio, options, service) {
+async function transcribeWithSpeech(audio, options, service, signal) {
+  if (signal?.aborted) throw createAbortError()
   if (!service) {
     throw new Error('DSH 自带语音输入未启用：请先在 插件 → Voice Input 中启用并准备本地模型，或配置 VoiceSense backendRoot')
   }
@@ -528,13 +623,32 @@ async function transcribeWithSpeech(audio, options, service) {
   const spec = service.resolve({
     audio,
     language: options.language || 'auto',
+    providerId: selected.id,
   })
-  const result = await service.transcribe(spec, AbortSignal.timeout(options.requestTimeoutMs))
+  const timeout = AbortSignal.timeout(options.requestTimeoutMs)
+  const combined = signal ? AbortSignal.any([signal, timeout]) : timeout
+  const result = await service.transcribe(spec, combined)
   const text = String(result?.text || '').trim()
   if (!text) throw new Error('DSH 语音输入没有返回文字')
   return text
 }
 
+function speechStatus(service) {
+  if (!service) return { ready: false, provider: null, phase: null }
+  try {
+    const snapshot = service.snapshot?.() ?? { providers: [], selection: {} }
+    const selected = snapshot.providers?.find((provider) => provider.id === snapshot.selection?.providerId)
+      ?? snapshot.providers?.[0]
+    return {
+      ready: Boolean(selected && speechProviderReady(selected)),
+      provider: selected?.id ?? null,
+      phase: selected?.preparation?.phase ?? null,
+    }
+  } catch {
+    return { ready: false, provider: null, phase: null }
+  }
+}
+
 function createAnalyzeHandler(options, backend, getSpeechToText) {
   return async (req, res) => {
     if (!isLocalRequest(req)) {
@@ -546,21 +660,37 @@ function createAnalyzeHandler(options, backend, getSpeechToText) {
       return
     }
     let tempDir
+    let controller
+    const onClose = () => {
+      if (res.writableEnded) return
+      controller?.abort()
+      backend.cancel()
+    }
     try {
-      const body = await readJsonBody(req, options.maxAudioBytes + 64 * 1024)
+      controller = new AbortController()
+      req.once?.('aborted', onClose)
+      res.once?.('close', onClose)
+      const bodyLimit = Math.ceil(options.maxAudioBytes / 3) * 4 + 64 * 1024
+      const body = await readJsonBody(req, bodyLimit)
+      if (!body || typeof body !== 'object' || Array.isArray(body)) {
+        throw new HttpError(400, '请求体必须是 JSON 对象')
+      }
       const base64 = String(body.audioBase64 || '').replace(/^data:[^,]+,/, '')
-      if (!base64) throw new Error('缺少录音数据')
+      if (!base64) throw new HttpError(400, '缺少录音数据')
       const buffer = Buffer.from(base64, 'base64')
-      if (buffer.length < 400) throw new Error('录音太短，没有听清')
-      if (buffer.length > options.maxAudioBytes) throw new Error('录音太长，请控制在 30 秒以内')
+      if (buffer.length < 400) throw new HttpError(400, '录音太短，没有听清')
+      if (buffer.length > options.maxAudioBytes) throw new HttpError(413, '录音太大，请缩短录音后重试')
       if (buffer.toString('ascii', 0, 4) !== 'RIFF' || buffer.toString('ascii', 8, 12) !== 'WAVE') {
-        throw new Error('录音格式不是 WAV')
+        throw new HttpError(400, '录音格式不是 WAV')
       }
       tempDir = await mkdtemp(join(tmpdir(), 'dsh-dafeiyu-emotion-'))
       const wavPath = join(tempDir, 'voice.wav')
       await writeFile(wavPath, buffer)
       let analysis
       let localError
+      if (options.backendRoot && !existsSync(options.backendRoot)) {
+        localError = `语音后端目录不存在：${options.backendRoot}`
+      }
       if (options.backendRoot && existsSync(options.backendRoot)) {
         try {
           analysis = await backend.analyze(wavPath.replaceAll('\\', '/'))
@@ -568,9 +698,10 @@ function createAnalyzeHandler(options, backend, getSpeechToText) {
           localError = error instanceof Error ? error.message : String(error)
         }
       }
+      if (controller.signal.aborted) return
       if (!analysis) {
         try {
-          const transcript = await transcribeWithSpeech(buffer, options, getSpeechToText())
+          const transcript = await transcribeWithSpeech(buffer, options, getSpeechToText(), controller.signal)
           analysis = decorateAnalysis(buildFallbackAnalysis(transcript, body.prosody || {}))
         } catch (speechError) {
           const parts = []
@@ -579,13 +710,18 @@ function createAnalyzeHandler(options, backend, getSpeechToText) {
           throw new Error(parts.join('\n'))
         }
       }
+      if (controller.signal.aborted || res.writableEnded) return
       jsonResponse(res, 200, analysis)
     } catch (error) {
-      jsonResponse(res, 500, {
+      if (controller?.signal.aborted || res.writableEnded) return
+      const status = error instanceof HttpError ? error.status : 500
+      jsonResponse(res, status, {
         error: error instanceof Error ? error.message : String(error),
-        detail: backend.stderr?.slice(-8).join('\n') || undefined,
+        detail: status >= 500 ? backend.stderr?.slice(-8).join('\n') || undefined : undefined,
       })
     } finally {
+      req.off?.('aborted', onClose)
+      res.off?.('close', onClose)
       if (tempDir) await rm(tempDir, { recursive: true, force: true }).catch(() => {})
     }
   }
@@ -596,6 +732,9 @@ export function apply(ctx, rawConfig = {}) {
   if (!options.enabled) return
   const logger = ctx.logger ?? console
   const backend = new VoiceBackend(options, logger)
+  if (options.backendRoot && !existsSync(options.backendRoot)) {
+    backend.noteError(`语音后端目录不存在：${options.backendRoot}`)
+  }
   let speechToText
   ctx.inject(['speechToText'], (speechCtx) => {
     speechToText = speechCtx.speechToText
@@ -615,13 +754,27 @@ export function apply(ctx, rawConfig = {}) {
             jsonResponse(res, 403, { error: '只允许本机访问' })
             return
           }
+          const speech = speechStatus(speechToText)
+          const backendConfigured = Boolean(options.backendRoot)
+          const backendAvailable = backendConfigured && existsSync(options.backendRoot)
+          if (backendAvailable && backend.lastError?.startsWith('语音后端目录不存在：')) {
+            backend.noteError(null)
+          }
           jsonResponse(res, 200, {
             ok: true,
             enabled: options.enabled,
             language: options.language,
-            engine: 'SenseVoice + emotion2vec',
-            mode: options.backendRoot ? 'enhanced-local-backend' : 'dsh-speech-prosody',
-            dshSpeechReady: Boolean(speechToText),
+            engine: backendAvailable ? 'SenseVoice + emotion2vec' : 'DSH Speech + prosody',
+            mode: !backendConfigured
+              ? 'dsh-speech-prosody'
+              : backendAvailable
+                ? 'enhanced-local-backend'
+                : 'enhanced-backend-missing',
+            dshSpeechReady: speech.ready,
+            speechProvider: speech.provider,
+            speechPhase: speech.phase,
+            backendConfigured,
+            backendAvailable,
             ...backend.status(),
           })
         },