feat: 实现语音转文字并支持流式输出；

fix: 修复了语音解密失败的问题

electron/main.ts

@@ -439,12 +439,14 @@ function registerIpcHandlers() {
     return chatService.getImageData(sessionId, msgId)
   })
 
-  ipcMain.handle('chat:getVoiceData', async (_, sessionId: string, msgId: string) => {
-    return chatService.getVoiceData(sessionId, msgId)
+  ipcMain.handle('chat:getVoiceData', async (_, sessionId: string, msgId: string, createTime?: number, serverId?: string | number) => {
+    return chatService.getVoiceData(sessionId, msgId, createTime, serverId)
   })
 
-  ipcMain.handle('chat:getVoiceTranscript', async (_, sessionId: string, msgId: string) => {
-    return chatService.getVoiceTranscript(sessionId, msgId)
+  ipcMain.handle('chat:getVoiceTranscript', async (event, sessionId: string, msgId: string) => {
+    return chatService.getVoiceTranscript(sessionId, msgId, (text) => {
+      event.sender.send('chat:voiceTranscriptPartial', { msgId, text })
+    })
   })
 
   ipcMain.handle('chat:getMessageById', async (_, sessionId: string, localId: number) => {
@@ -521,14 +523,14 @@ function registerIpcHandlers() {
     return { success: true }
   })
 
-  ipcMain.handle('whisper:downloadModel', async (event, payload: { modelName: string; downloadDir?: string; source?: string }) => {
-    return voiceTranscribeService.downloadModel(payload, (progress) => {
+  ipcMain.handle('whisper:downloadModel', async (event) => {
+    return voiceTranscribeService.downloadModel((progress) => {
       event.sender.send('whisper:downloadProgress', progress)
     })
   })
 
-  ipcMain.handle('whisper:getModelStatus', async (_, payload: { modelName: string; downloadDir?: string }) => {
-    return voiceTranscribeService.getModelStatus(payload)
+  ipcMain.handle('whisper:getModelStatus', async () => {
+    return voiceTranscribeService.getModelStatus()
   })
 
   // 群聊分析相关

electron/preload.ts

@@ -106,8 +106,14 @@ contextBridge.exposeInMainWorld('electronAPI', {
     close: () => ipcRenderer.invoke('chat:close'),
     getSessionDetail: (sessionId: string) => ipcRenderer.invoke('chat:getSessionDetail', sessionId),
     getImageData: (sessionId: string, msgId: string) => ipcRenderer.invoke('chat:getImageData', sessionId, msgId),
-    getVoiceData: (sessionId: string, msgId: string) => ipcRenderer.invoke('chat:getVoiceData', sessionId, msgId),
-    getVoiceTranscript: (sessionId: string, msgId: string) => ipcRenderer.invoke('chat:getVoiceTranscript', sessionId, msgId)
+    getVoiceData: (sessionId: string, msgId: string, createTime?: number, serverId?: string | number) =>
+      ipcRenderer.invoke('chat:getVoiceData', sessionId, msgId, createTime, serverId),
+    getVoiceTranscript: (sessionId: string, msgId: string) => ipcRenderer.invoke('chat:getVoiceTranscript', sessionId, msgId),
+    onVoiceTranscriptPartial: (callback: (payload: { msgId: string; text: string }) => void) => {
+      const listener = (_: any, payload: { msgId: string; text: string }) => callback(payload)
+      ipcRenderer.on('chat:voiceTranscriptPartial', listener)
+      return () => ipcRenderer.removeListener('chat:voiceTranscriptPartial', listener)
+    }
   },
 
 

electron/services/analyticsService.ts

@@ -324,7 +324,7 @@ class AnalyticsService {
   }
 
   private getCacheFilePath(): string {
-    return join(app.getPath('userData'), 'analytics_cache.json')
+    return join(app.getPath('documents'), 'WeFlow', 'analytics_cache.json')
   }
 
   private async loadCacheFromFile(): Promise<{ key: string; data: any; updatedAt: number } | null> {

electron/services/chatService.ts

@@ -7,11 +7,7 @@ import * as http from 'http'
 import * as fzstd from 'fzstd'
 import * as crypto from 'crypto'
 import Database from 'better-sqlite3'
-import { execFile } from 'child_process'
-import { promisify } from 'util'
 import { app } from 'electron'
-
-const execFileAsync = promisify(execFile)
 import { ConfigService } from './config'
 import { wcdbService } from './wcdbService'
 import { MessageCacheService } from './messageCacheService'
@@ -2149,7 +2145,107 @@ class ChatService {
     }
   }
 
-  async getVoiceData(sessionId: string, msgId: string): Promise<{ success: boolean; data?: string; error?: string }> {
+  /**
+   * getVoiceData (优化的 C++ 实现 + 文件缓存)
+   */
+  async getVoiceData(sessionId: string, msgId: string, createTime?: number, serverId?: string | number): Promise<{ success: boolean; data?: string; error?: string }> {
+
+    try {
+      const localId = parseInt(msgId, 10)
+      if (isNaN(localId)) {
+        return { success: false, error: '无效的消息ID' }
+      }
+
+      // 检查文件缓存
+      const cacheKey = this.getVoiceCacheKey(sessionId, msgId)
+      const cachedFile = this.getVoiceCacheFilePath(cacheKey)
+      if (existsSync(cachedFile)) {
+        try {
+          const wavData = readFileSync(cachedFile)
+          console.info('[ChatService][Voice] 使用缓存文件:', cachedFile)
+          return { success: true, data: wavData.toString('base64') }
+        } catch (e) {
+          console.error('[ChatService][Voice] 读取缓存失败:', e)
+          // 继续重新解密
+        }
+      }
+
+      // 1. 确定 createTime 和 svrId
+      let msgCreateTime = createTime
+      let msgSvrId: string | number = serverId || 0
+
+      // 如果提供了传来的参数，验证其有效性
+      if (!msgCreateTime || msgCreateTime === 0) {
+        const msgResult = await this.getMessageByLocalId(sessionId, localId)
+        if (msgResult.success && msgResult.message) {
+          const msg = msgResult.message as any
+          msgCreateTime = msg.createTime || msg.create_time
+          // 尝试获取各种可能的 server id 列名 (只有在没有传入 serverId 时才查找)
+          if (!msgSvrId || msgSvrId === 0) {
+            msgSvrId = msg.serverId || msg.svr_id || msg.msg_svr_id || msg.message_id || 0
+          }
+        }
+      }
+
+      if (!msgCreateTime) {
+        return { success: false, error: '未找到消息时间戳' }
+      }
+
+      // 2. 构建查找候选 (sessionId, myWxid)
+      const candidates: string[] = []
+      if (sessionId) candidates.push(sessionId)
+      const myWxid = this.configService.get('myWxid') as string
+      if (myWxid && !candidates.includes(myWxid)) {
+        candidates.push(myWxid)
+      }
+
+
+
+      // 3. 调用 C++ 接口获取语音 (Hex)
+      const voiceRes = await wcdbService.getVoiceData(sessionId, msgCreateTime, candidates, msgSvrId)
+      if (!voiceRes.success || !voiceRes.hex) {
+        return { success: false, error: voiceRes.error || '未找到语音数据' }
+      }
+
+
+
+      // 4. Hex 转 Buffer (Silk)
+      const silkData = Buffer.from(voiceRes.hex, 'hex')
+
+      // 5. 使用 silk-wasm 解码
+      try {
+        const pcmData = await this.decodeSilkToPcm(silkData, 24000)
+        if (!pcmData) {
+          return { success: false, error: 'Silk 解码失败' }
+        }
+
+        // PCM -> WAV
+        const wavData = this.createWavBuffer(pcmData, 24000)
+
+        // 保存到文件缓存
+        try {
+          this.saveVoiceCache(cacheKey, wavData)
+          console.info('[ChatService][Voice] 已保存缓存:', cachedFile)
+        } catch (e) {
+          console.error('[ChatService][Voice] 保存缓存失败:', e)
+          // 不影响返回
+        }
+
+        // 缓存 WAV 数据 (内存缓存)
+        this.cacheVoiceWav(cacheKey, wavData)
+
+        return { success: true, data: wavData.toString('base64') }
+      } catch (e) {
+        console.error('[ChatService][Voice] decoding error:', e)
+        return { success: false, error: '语音解码失败: ' + String(e) }
+      }
+    } catch (e) {
+      console.error('ChatService: getVoiceData 失败:', e)
+      return { success: false, error: String(e) }
+    }
+  }
+
+  async getVoiceData_Legacy(sessionId: string, msgId: string): Promise<{ success: boolean; data?: string; error?: string }> {
     try {
       const localId = parseInt(msgId, 10)
       const msgResult = await this.getMessageByLocalId(sessionId, localId)
@@ -2187,12 +2283,10 @@ class ChatService {
       for (const dbPath of (mediaDbs.data || [])) {
         const voiceTable = await this.resolveVoiceInfoTableName(dbPath)
         if (!voiceTable) {
-          console.warn('[ChatService][Voice] voice table not found', dbPath)
           continue
         }
         const columns = await this.resolveVoiceInfoColumns(dbPath, voiceTable)
         if (!columns) {
-          console.warn('[ChatService][Voice] voice columns not found', { dbPath, voiceTable })
           continue
         }
         for (const candidate of candidates) {
@@ -2233,52 +2327,44 @@ class ChatService {
           }
         }
         if (silkData) break
+
+        // 策略 3: 只使用 CreateTime (兜底)
+        if (!silkData && columns.createTimeColumn) {
+          const whereClause = `${columns.createTimeColumn} = ${msg.createTime}`
+          const sql = `SELECT ${columns.dataColumn} AS data FROM ${voiceTable} WHERE ${whereClause} LIMIT 1`
+          const result = await wcdbService.execQuery('media', dbPath, sql)
+          if (result.success && result.rows && result.rows.length > 0) {
+            const raw = result.rows[0]?.data
+            const decoded = this.decodeVoiceBlob(raw)
+            if (decoded && decoded.length > 0) {
+              console.info('[ChatService][Voice] hit by createTime only', { dbPath, voiceTable, whereClause, bytes: decoded.length })
+              silkData = decoded
+            }
+          }
+        }
+        if (silkData) break
       }
 
       if (!silkData) return { success: false, error: '未找到语音数据' }
 
-      // 4. 解码 Silk -> PCM -> WAV
-      const resourcesPath = app.isPackaged
-        ? join(process.resourcesPath, 'resources')
-        : join(app.getAppPath(), 'resources')
-      const decoderPath = join(resourcesPath, 'silk_v3_decoder.exe')
-
-      if (!existsSync(decoderPath)) {
-        return { success: false, error: '找不到语音解码器 (silk_v3_decoder.exe)' }
-      }
-      console.info('[ChatService][Voice] decoder path', decoderPath)
-
-      const tempDir = app.getPath('temp')
-      const silkFile = join(tempDir, `voice_${msgId}.silk`)
-      const pcmFile = join(tempDir, `voice_${msgId}.pcm`)
-
+      // 4. 使用 silk-wasm 解码
       try {
-        writeFileSync(silkFile, silkData)
-        // 执行解码: silk_v3_decoder.exe <silk> <pcm> -Fs_API 24000
-        console.info('[ChatService][Voice] executing decoder:', decoderPath, [silkFile, pcmFile])
-        const { stdout, stderr } = await execFileAsync(
-          decoderPath,
-          [silkFile, pcmFile, '-Fs_API', '24000'],
-          { cwd: dirname(decoderPath) }
-        )
-        if (stdout && stdout.trim()) console.info('[ChatService][Voice] decoder stdout:', stdout)
-        if (stderr && stderr.trim()) console.warn('[ChatService][Voice] decoder stderr:', stderr)
-
-        if (!existsSync(pcmFile)) {
-          return { success: false, error: '语音解码失败' }
+        const pcmData = await this.decodeSilkToPcm(silkData, 24000)
+        if (!pcmData) {
+          return { success: false, error: 'Silk 解码失败' }
         }
 
-        const pcmData = readFileSync(pcmFile)
-        const wavHeader = this.createWavHeader(pcmData.length, 24000, 1) // 微信语音通常 24kHz
-        const wavData = Buffer.concat([wavHeader, pcmData])
+        // PCM -> WAV
+        const wavData = this.createWavBuffer(pcmData, 24000)
+
+        // 缓存 WAV 数据 (内存缓存)
         const cacheKey = this.getVoiceCacheKey(sessionId, msgId)
         this.cacheVoiceWav(cacheKey, wavData)
 
         return { success: true, data: wavData.toString('base64') }
-      } finally {
-        // 清理临时文件
-        try { if (existsSync(silkFile)) unlinkSync(silkFile) } catch { }
-        try { if (existsSync(pcmFile)) unlinkSync(pcmFile) } catch { }
+      } catch (e) {
+        console.error('[ChatService][Voice] decoding error:', e)
+        return { success: false, error: '语音解码失败: ' + String(e) }
       }
     } catch (e) {
       console.error('ChatService: getVoiceData 失败:', e)
@@ -2286,7 +2372,69 @@ class ChatService {
     }
   }
 
-  async getVoiceTranscript(sessionId: string, msgId: string): Promise<{ success: boolean; transcript?: string; error?: string }> {
+
+
+  /**
+   * 解码 Silk 数据为 PCM (silk-wasm)
+   */
+  private async decodeSilkToPcm(silkData: Buffer, sampleRate: number): Promise<Buffer | null> {
+    try {
+      let wasmPath: string
+      if (app.isPackaged) {
+        wasmPath = join(process.resourcesPath, 'app.asar.unpacked', 'node_modules', 'silk-wasm', 'lib', 'silk.wasm')
+        if (!existsSync(wasmPath)) {
+          wasmPath = join(process.resourcesPath, 'node_modules', 'silk-wasm', 'lib', 'silk.wasm')
+        }
+      } else {
+        wasmPath = join(app.getAppPath(), 'node_modules', 'silk-wasm', 'lib', 'silk.wasm')
+      }
+
+      if (!existsSync(wasmPath)) {
+        console.error('[ChatService][Voice] silk.wasm not found at:', wasmPath)
+        return null
+      }
+
+      const silkWasm = require('silk-wasm')
+      if (!silkWasm || !silkWasm.decode) {
+        console.error('[ChatService][Voice] silk-wasm module invalid')
+        return null
+      }
+
+      const result = await silkWasm.decode(silkData, sampleRate)
+      return Buffer.from(result.data)
+    } catch (e) {
+      console.error('[ChatService][Voice] internal decode error:', e)
+      return null
+    }
+  }
+
+  /**
+   * 创建 WAV 文件 Buffer
+   */
+  private createWavBuffer(pcmData: Buffer, sampleRate: number = 24000, channels: number = 1): Buffer {
+    const pcmLength = pcmData.length
+    const header = Buffer.alloc(44)
+    header.write('RIFF', 0)
+    header.writeUInt32LE(36 + pcmLength, 4)
+    header.write('WAVE', 8)
+    header.write('fmt ', 12)
+    header.writeUInt32LE(16, 16)
+    header.writeUInt16LE(1, 20)
+    header.writeUInt16LE(channels, 22)
+    header.writeUInt32LE(sampleRate, 24)
+    header.writeUInt32LE(sampleRate * channels * 2, 28)
+    header.writeUInt16LE(channels * 2, 32)
+    header.writeUInt16LE(16, 34)
+    header.write('data', 36)
+    header.writeUInt32LE(pcmLength, 40)
+    return Buffer.concat([header, pcmData])
+  }
+
+  async getVoiceTranscript(
+    sessionId: string,
+    msgId: string,
+    onPartial?: (text: string) => void
+  ): Promise<{ success: boolean; transcript?: string; error?: string }> {
     const cacheKey = this.getVoiceCacheKey(sessionId, msgId)
     const cached = this.voiceTranscriptCache.get(cacheKey)
     if (cached) {
@@ -2302,14 +2450,25 @@ class ChatService {
       try {
         let wavData = this.voiceWavCache.get(cacheKey)
         if (!wavData) {
-          const voiceResult = await this.getVoiceData(sessionId, msgId)
+          // 获取消息详情以拿到 createTime 和 serverId
+          let cTime: number | undefined
+          let sId: string | number | undefined
+          const msgResult = await this.getMessageById(sessionId, parseInt(msgId, 10))
+          if (msgResult.success && msgResult.message) {
+            cTime = msgResult.message.createTime
+            sId = msgResult.message.serverId
+          }
+
+          const voiceResult = await this.getVoiceData(sessionId, msgId, cTime, sId)
           if (!voiceResult.success || !voiceResult.data) {
             return { success: false, error: voiceResult.error || '语音解码失败' }
           }
           wavData = Buffer.from(voiceResult.data, 'base64')
         }
 
-        const result = await voiceTranscribeService.transcribeWavBuffer(wavData)
+        const result = await voiceTranscribeService.transcribeWavBuffer(wavData, (text) => {
+          onPartial?.(text)
+        })
         if (result.success && result.transcript) {
           this.cacheVoiceTranscript(cacheKey, result.transcript)
         }
@@ -2325,26 +2484,10 @@ class ChatService {
     return task
   }
 
-  private createWavHeader(pcmLength: number, sampleRate: number = 24000, channels: number = 1): Buffer {
-    const header = Buffer.alloc(44)
-    header.write('RIFF', 0)
-    header.writeUInt32LE(36 + pcmLength, 4)
-    header.write('WAVE', 8)
-    header.write('fmt ', 12)
-    header.writeUInt32LE(16, 16)
-    header.writeUInt16LE(1, 20)
-    header.writeUInt16LE(channels, 22)
-    header.writeUInt32LE(sampleRate, 24)
-    header.writeUInt32LE(sampleRate * channels * 2, 28)
-    header.writeUInt16LE(channels * 2, 32)
-    header.writeUInt16LE(16, 34)
-    header.write('data', 36)
-    header.writeUInt32LE(pcmLength, 40)
-    return header
-  }
+
 
   private getVoiceCacheKey(sessionId: string, msgId: string): string {
-    return `${sessionId}:${msgId}`
+    return `${sessionId}_${msgId}`
   }
 
   private cacheVoiceWav(cacheKey: string, wavData: Buffer): void {
@@ -2355,6 +2498,32 @@ class ChatService {
     }
   }
 
+  /**
+   * 获取语音缓存文件路径
+   */
+  private getVoiceCacheFilePath(cacheKey: string): string {
+    const cachePath = this.configService.get('cachePath') as string | undefined
+    let baseDir: string
+    if (cachePath && cachePath.trim()) {
+      baseDir = join(cachePath, 'Voices')
+    } else {
+      const documentsPath = app.getPath('documents')
+      baseDir = join(documentsPath, 'WeFlow', 'Voices')
+    }
+    if (!existsSync(baseDir)) {
+      mkdirSync(baseDir, { recursive: true })
+    }
+    return join(baseDir, `${cacheKey}.wav`)
+  }
+
+  /**
+   * 保存语音到文件缓存
+   */
+  private saveVoiceCache(cacheKey: string, wavData: Buffer): void {
+    const filePath = this.getVoiceCacheFilePath(cacheKey)
+    writeFileSync(filePath, wavData)
+  }
+
   private cacheVoiceTranscript(cacheKey: string, transcript: string): void {
     this.voiceTranscriptCache.set(cacheKey, transcript)
     if (this.voiceTranscriptCache.size > this.voiceCacheMaxEntries) {

electron/services/contactCacheService.ts

@@ -15,7 +15,7 @@ export class ContactCacheService {
   constructor(cacheBasePath?: string) {
     const basePath = cacheBasePath && cacheBasePath.trim().length > 0
       ? cacheBasePath
-      : join(app.getPath('userData'), 'WeFlowCache')
+      : join(app.getPath('documents'), 'WeFlow')
     this.cacheFilePath = join(basePath, 'contacts.json')
     this.ensureCacheDir()
     this.loadCache()

electron/services/exportService.ts

@@ -70,6 +70,7 @@ export interface ExportOptions {
   exportImages?: boolean
   exportVoices?: boolean
   exportEmojis?: boolean
+  exportVoiceAsText?: boolean
 }
 
 interface MediaExportItem {
@@ -227,6 +228,7 @@ class ExportService {
 
   /**
    * 解析消息内容为可读文本
+   * 注意：语音消息在这里返回占位符，实际转文字在导出时异步处理
    */
   private parseMessageContent(content: string, localType: number): string | null {
     if (!content) return null
@@ -235,7 +237,7 @@ class ExportService {
       case 1:
         return this.stripSenderPrefix(content)
       case 3: return '[图片]'
-      case 34: return '[语音消息]'
+      case 34: return '[语音消息]'  // 占位符，导出时会替换为转文字结果
       case 42: return '[名片]'
       case 43: return '[视频]'
       case 47: return '[动画表情]'
@@ -246,6 +248,7 @@ class ExportService {
       }
       case 50: return this.parseVoipMessage(content)
       case 10000: return this.cleanSystemMessage(content)
+      case 266287972401: return this.cleanSystemMessage(content)  // 拍一拍
       default:
         if (content.includes('<type>57</type>')) {
           const title = this.extractXmlValue(content, 'title')
@@ -270,20 +273,20 @@ class ExportService {
 
   private cleanSystemMessage(content: string): string {
     if (!content) return '[系统消息]'
-    
+
     // 先尝试提取特定的系统消息内容
     // 1. 提取 sysmsg 中的文本内容
     const sysmsgTextMatch = /<sysmsg[^>]*>([\s\S]*?)<\/sysmsg>/i.exec(content)
     if (sysmsgTextMatch) {
       content = sysmsgTextMatch[1]
     }
-    
+
     // 2. 提取 revokemsg 撤回消息
     const revokeMatch = /<replacemsg><!\[CDATA\[(.*?)\]\]><\/replacemsg>/i.exec(content)
     if (revokeMatch) {
       return revokeMatch[1].trim()
     }
-    
+
     // 3. 提取 pat 拍一拍消息
     const patMatch = /<template><!\[CDATA\[(.*?)\]\]><\/template>/i.exec(content)
     if (patMatch) {
@@ -296,10 +299,10 @@ class ExportService {
         .replace(/<[^>]+>/g, '')
         .trim()
     }
-    
+
     // 4. 处理 CDATA 内容
     content = content.replace(/<!\[CDATA\[/g, '').replace(/\]\]>/g, '')
-    
+
     // 5. 移除所有 XML 标签
     return content
       .replace(/<img[^>]*>/gi, '')
@@ -406,10 +409,10 @@ class ExportService {
     msg: any,
     sessionId: string,
     mediaDir: string,
-    options: { exportImages?: boolean; exportVoices?: boolean; exportEmojis?: boolean }
+    options: { exportImages?: boolean; exportVoices?: boolean; exportEmojis?: boolean; exportVoiceAsText?: boolean }
   ): Promise<MediaExportItem | null> {
     const localType = msg.localType
-    
+
     // 图片消息
     if (localType === 3 && options.exportImages) {
       const result = await this.exportImage(msg, sessionId, mediaDir)
@@ -418,12 +421,19 @@ class ExportService {
       }
       return result
     }
-    
+
     // 语音消息
-    if (localType === 34 && options.exportVoices) {
-      return this.exportVoice(msg, sessionId, mediaDir)
+    if (localType === 34) {
+      // 如果开启了语音转文字，优先转文字（不导出语音文件）
+      if (options.exportVoiceAsText) {
+        return null  // 转文字逻辑在消息内容处理中完成
+      }
+      // 否则导出语音文件
+      if (options.exportVoices) {
+        return this.exportVoice(msg, sessionId, mediaDir)
+      }
     }
-    
+
     // 动画表情
     if (localType === 47 && options.exportEmojis) {
       const result = await this.exportEmoji(msg, sessionId, mediaDir)
@@ -432,7 +442,7 @@ class ExportService {
       }
       return result
     }
-    
+
     return null
   }
 
@@ -449,7 +459,7 @@ class ExportService {
       // 使用消息对象中已提取的字段
       const imageMd5 = msg.imageMd5
       const imageDatName = msg.imageDatName
-      
+
       if (!imageMd5 && !imageDatName) {
         console.log('[ExportService] 图片消息缺少 md5 和 datName:', msg.localId)
         return null
@@ -485,9 +495,9 @@ class ExportService {
         const ext = this.getExtFromDataUrl(sourcePath)
         const fileName = `${imageMd5 || imageDatName || msg.localId}${ext}`
         const destPath = path.join(imagesDir, fileName)
-        
+
         fs.writeFileSync(destPath, Buffer.from(base64Data, 'base64'))
-        
+
         return {
           relativePath: `media/images/${fileName}`,
           kind: 'image'
@@ -501,11 +511,11 @@ class ExportService {
         const ext = path.extname(sourcePath) || '.jpg'
         const fileName = `${imageMd5 || imageDatName || msg.localId}${ext}`
         const destPath = path.join(imagesDir, fileName)
-        
+
         if (!fs.existsSync(destPath)) {
           fs.copyFileSync(sourcePath, destPath)
         }
-        
+
         return {
           relativePath: `media/images/${fileName}`,
           kind: 'image'
@@ -566,6 +576,22 @@ class ExportService {
     }
   }
 
+  /**
+   * 转写语音为文字
+   */
+  private async transcribeVoice(sessionId: string, msgId: string): Promise<string> {
+    try {
+      const transcript = await chatService.getVoiceTranscript(sessionId, msgId)
+      if (transcript.success && transcript.transcript) {
+        return `[语音转文字] ${transcript.transcript}`
+      }
+      return '[语音消息 - 转文字失败]'
+    } catch (e) {
+      console.error('[ExportService] 语音转文字失败:', e)
+      return '[语音消息 - 转文字失败]'
+    }
+  }
+
   /**
    * 导出表情文件
    */
@@ -579,7 +605,7 @@ class ExportService {
       // 使用消息对象中已提取的字段
       const emojiUrl = msg.emojiCdnUrl
       const emojiMd5 = msg.emojiMd5
-      
+
       if (!emojiUrl && !emojiMd5) {
         console.log('[ExportService] 表情消息缺少 url 和 md5, localId:', msg.localId, 'content:', msg.content?.substring(0, 200))
         return null
@@ -669,7 +695,7 @@ class ExportService {
         if (url.includes('%')) {
           url = decodeURIComponent(url)
         }
-      } catch {}
+      } catch { }
       return url
     }
     // 备用：尝试 XML 标签形式
@@ -792,7 +818,7 @@ class ExportService {
           let imageDatName: string | undefined
           let emojiCdnUrl: string | undefined
           let emojiMd5: string | undefined
-          
+
           if (localType === 3 && content) {
             // 图片消息
             imageMd5 = this.extractImageMd5(content)
@@ -1057,6 +1083,31 @@ class ExportService {
     }
   }
 
+  /**
+   * 生成通用的导出元数据 (参考 ChatLab 格式)
+   */
+  private getExportMeta(
+    sessionId: string,
+    sessionInfo: { displayName: string },
+    isGroup: boolean,
+    sessionAvatar?: string
+  ): { chatlab: ChatLabHeader; meta: ChatLabMeta } {
+    return {
+      chatlab: {
+        version: '0.0.2',
+        exportedAt: Math.floor(Date.now() / 1000),
+        generator: 'WeFlow'
+      },
+      meta: {
+        name: sessionInfo.displayName,
+        platform: 'wechat',
+        type: isGroup ? 'group' : 'private',
+        ...(isGroup && { groupId: sessionId }),
+        ...(sessionAvatar && { groupAvatar: sessionAvatar })
+      }
+    }
+  }
+
   /**
    * 导出单个会话为 ChatLab 格式
    */
@@ -1097,21 +1148,29 @@ class ExportService {
         phase: 'exporting'
       })
 
-      const chatLabMessages: ChatLabMessage[] = allMessages.map((msg) => {
+      const chatLabMessages: ChatLabMessage[] = []
+      for (const msg of allMessages) {
         const memberInfo = collected.memberSet.get(msg.senderUsername)?.member || {
           platformId: msg.senderUsername,
           accountName: msg.senderUsername,
           groupNickname: undefined
         }
-        return {
+
+        let content = this.parseMessageContent(msg.content, msg.localType)
+        // 如果是语音消息且开启了转文字
+        if (msg.localType === 34 && options.exportVoiceAsText) {
+          content = await this.transcribeVoice(sessionId, String(msg.localId))
+        }
+
+        chatLabMessages.push({
           sender: msg.senderUsername,
           accountName: memberInfo.accountName,
           groupNickname: memberInfo.groupNickname,
           timestamp: msg.createTime,
           type: this.convertMessageType(msg.localType, msg.content),
-          content: this.parseMessageContent(msg.content, msg.localType)
-        }
-      })
+          content: content
+        })
+      }
 
       const avatarMap = options.exportAvatars
         ? await this.exportAvatars(
@@ -1131,19 +1190,11 @@ class ExportService {
         return avatar ? { ...info.member, avatar } : info.member
       })
 
+      const { chatlab, meta } = this.getExportMeta(sessionId, sessionInfo, isGroup, sessionAvatar)
+
       const chatLabExport: ChatLabExport = {
-        chatlab: {
-          version: '0.0.1',
-          exportedAt: Math.floor(Date.now() / 1000),
-          generator: 'WeFlow'
-        },
-        meta: {
-          name: sessionInfo.displayName,
-          platform: 'wechat',
-          type: isGroup ? 'group' : 'private',
-          ...(isGroup && { groupId: sessionId }),
-          ...(sessionAvatar && { groupAvatar: sessionAvatar })
-        },
+        chatlab,
+        meta,
         members,
         messages: chatLabMessages
       }
@@ -1245,7 +1296,11 @@ class ExportService {
         phase: 'writing'
       })
 
-      const detailedExport = {
+      const { chatlab, meta } = this.getExportMeta(sessionId, sessionInfo, isGroup)
+
+      const detailedExport: any = {
+        chatlab,
+        meta,
         session: {
           wxid: sessionId,
           nickname: sessionInfo.displayName,
@@ -1316,7 +1371,7 @@ class ExportService {
 
       const sessionInfo = await this.getContactInfo(sessionId)
       const myInfo = await this.getContactInfo(cleanedMyWxid)
-      
+
       // 获取会话的备注信息
       const sessionContact = await wcdbService.getContact(sessionId)
       const sessionRemark = sessionContact.success && sessionContact.contact?.remark ? sessionContact.contact.remark : ''
@@ -1362,12 +1417,12 @@ class ExportService {
       worksheet.mergeCells(currentRow, 2, currentRow, 3)
       worksheet.getCell(currentRow, 2).value = sessionId
       worksheet.getCell(currentRow, 2).font = { name: 'Calibri', size: 11 }
-      
+
       worksheet.getCell(currentRow, 4).value = '昵称'
       worksheet.getCell(currentRow, 4).font = { name: 'Calibri', bold: true, size: 11 }
       worksheet.getCell(currentRow, 5).value = sessionNickname
       worksheet.getCell(currentRow, 5).font = { name: 'Calibri', size: 11 }
-      
+
       if (isGroup) {
         worksheet.getCell(currentRow, 6).value = '备注'
         worksheet.getCell(currentRow, 6).font = { name: 'Calibri', bold: true, size: 11 }
@@ -1378,11 +1433,36 @@ class ExportService {
       worksheet.getRow(currentRow).height = 20
       currentRow++
 
+      // 第三行：导出元数据
+      const { chatlab, meta: exportMeta } = this.getExportMeta(sessionId, sessionInfo, isGroup)
+      worksheet.getCell(currentRow, 1).value = '导出工具'
+      worksheet.getCell(currentRow, 1).font = { name: 'Calibri', bold: true, size: 11 }
+      worksheet.getCell(currentRow, 2).value = chatlab.generator
+      worksheet.getCell(currentRow, 2).font = { name: 'Calibri', size: 10 }
+
+      worksheet.getCell(currentRow, 3).value = '导出版本'
+      worksheet.getCell(currentRow, 3).font = { name: 'Calibri', bold: true, size: 11 }
+      worksheet.getCell(currentRow, 4).value = chatlab.version
+      worksheet.getCell(currentRow, 4).font = { name: 'Calibri', size: 10 }
+
+      worksheet.getCell(currentRow, 5).value = '平台'
+      worksheet.getCell(currentRow, 5).font = { name: 'Calibri', bold: true, size: 11 }
+      worksheet.getCell(currentRow, 6).value = exportMeta.platform
+      worksheet.getCell(currentRow, 6).font = { name: 'Calibri', size: 10 }
+
+      worksheet.getCell(currentRow, 7).value = '导出时间'
+      worksheet.getCell(currentRow, 7).font = { name: 'Calibri', bold: true, size: 11 }
+      worksheet.getCell(currentRow, 8).value = this.formatTimestamp(chatlab.exportedAt)
+      worksheet.getCell(currentRow, 8).font = { name: 'Calibri', size: 10 }
+
+      worksheet.getRow(currentRow).height = 20
+      currentRow++
+
       // 表头行
       const headers = ['序号', '时间', '发送者昵称', '发送者微信ID', '发送者备注', '发送者身份', '消息类型', '内容']
       const headerRow = worksheet.getRow(currentRow)
       headerRow.height = 22
-      
+
       headers.forEach((header, index) => {
         const cell = headerRow.getCell(index + 1)
         cell.value = header
@@ -1408,17 +1488,17 @@ class ExportService {
 
       // 填充数据
       const sortedMessages = collected.rows.sort((a, b) => a.createTime - b.createTime)
-      
+
       // 媒体导出设置
       const exportMediaEnabled = options.exportImages || options.exportVoices || options.exportEmojis
       const sessionDir = path.dirname(outputPath)  // 会话目录，用于媒体导出
-      
+
       // 媒体导出缓存
       const mediaCache = new Map<string, MediaExportItem | null>()
-      
+
       for (let i = 0; i < sortedMessages.length; i++) {
         const msg = sortedMessages[i]
-        
+
         // 导出媒体文件
         let mediaItem: MediaExportItem | null = null
         if (exportMediaEnabled) {
@@ -1429,18 +1509,19 @@ class ExportService {
             mediaItem = await this.exportMediaForMessage(msg, sessionId, sessionDir, {
               exportImages: options.exportImages,
               exportVoices: options.exportVoices,
-              exportEmojis: options.exportEmojis
+              exportEmojis: options.exportEmojis,
+              exportVoiceAsText: options.exportVoiceAsText
             })
             mediaCache.set(mediaKey, mediaItem)
           }
         }
-        
+
         // 确定发送者信息
         let senderRole: string
         let senderWxid: string
         let senderNickname: string
         let senderRemark: string = ''
-        
+
         if (msg.isSend) {
           // 我发送的消息
           senderRole = '我'
@@ -1450,7 +1531,7 @@ class ExportService {
         } else if (isGroup && msg.senderUsername) {
           // 群消息
           senderWxid = msg.senderUsername
-          
+
           // 用 getContact 获取联系人详情，分别取昵称和备注
           const contactDetail = await wcdbService.getContact(msg.senderUsername)
           if (contactDetail.success && contactDetail.contact) {
@@ -1481,12 +1562,12 @@ class ExportService {
 
         const row = worksheet.getRow(currentRow)
         row.height = 24
-        
+
         // 确定内容：如果有媒体文件导出成功则显示相对路径，否则显示解析后的内容
-        const contentValue = mediaItem 
-          ? mediaItem.relativePath 
+        const contentValue = mediaItem
+          ? mediaItem.relativePath
           : (this.parseMessageContent(msg.content, msg.localType) || '')
-        
+
         // 调试日志
         if (msg.localType === 3 || msg.localType === 47) {
           console.log('[ExportService] 媒体消息填充表格:', {
@@ -1497,7 +1578,7 @@ class ExportService {
             contentValue: contentValue?.substring(0, 100)
           })
         }
-        
+
         worksheet.getCell(currentRow, 1).value = i + 1
         worksheet.getCell(currentRow, 2).value = this.formatTimestamp(msg.createTime)
         worksheet.getCell(currentRow, 3).value = senderNickname
@@ -1506,14 +1587,14 @@ class ExportService {
         worksheet.getCell(currentRow, 6).value = senderRole
         worksheet.getCell(currentRow, 7).value = this.getMessageTypeName(msg.localType)
         worksheet.getCell(currentRow, 8).value = contentValue
-        
+
         // 设置每个单元格的样式
         for (let col = 1; col <= 8; col++) {
           const cell = worksheet.getCell(currentRow, col)
           cell.font = { name: 'Calibri', size: 11 }
           cell.alignment = { vertical: 'middle', wrapText: false }
         }
-        
+
         currentRow++
 
         // 每处理 100 条消息报告一次进度
@@ -1548,14 +1629,14 @@ class ExportService {
       return { success: true }
     } catch (e) {
       console.error('ExportService: 导出 Excel 失败:', e)
-      
+
       // 处理文件被占用的错误
       if (e instanceof Error) {
         if (e.message.includes('EBUSY') || e.message.includes('resource busy') || e.message.includes('locked')) {
           return { success: false, error: '文件已经打开，请关闭后再导出' }
         }
       }
-      
+
       return { success: false, error: String(e) }
     }
   }
@@ -1594,13 +1675,13 @@ class ExportService {
         })
 
         const safeName = sessionInfo.displayName.replace(/[<>:"/\\|?*]/g, '_')
-        
+
         // 为每个会话创建单独的文件夹
         const sessionDir = path.join(outputDir, safeName)
         if (!fs.existsSync(sessionDir)) {
           fs.mkdirSync(sessionDir, { recursive: true })
         }
-        
+
         let ext = '.json'
         if (options.format === 'chatlab-jsonl') ext = '.jsonl'
         else if (options.format === 'excel') ext = '.xlsx'

electron/services/imageDecryptService.ts

@@ -14,17 +14,17 @@ function getStaticFfmpegPath(): string | null {
     // 方法1: 直接 require ffmpeg-static
     // eslint-disable-next-line @typescript-eslint/no-var-requires
     const ffmpegStatic = require('ffmpeg-static')
-    
+
     if (typeof ffmpegStatic === 'string' && existsSync(ffmpegStatic)) {
       return ffmpegStatic
     }
-    
+
     // 方法2: 手动构建路径（开发环境）
     const devPath = join(process.cwd(), 'node_modules', 'ffmpeg-static', 'ffmpeg.exe')
     if (existsSync(devPath)) {
       return devPath
     }
-    
+
     // 方法3: 打包后的路径
     if (app.isPackaged) {
       const resourcesPath = process.resourcesPath
@@ -33,7 +33,7 @@ function getStaticFfmpegPath(): string | null {
         return packedPath
       }
     }
-    
+
     return null
   } catch {
     return null
@@ -115,7 +115,6 @@ export class ImageDecryptService {
     for (const key of cacheKeys) {
       const cached = this.resolvedCache.get(key)
       if (cached && existsSync(cached) && this.isImageFile(cached)) {
-        this.logInfo('缓存命中(从Map)', { key, path: cached, isThumb: this.isThumbnailPath(cached) })
         const dataUrl = this.fileToDataUrl(cached)
         const isThumb = this.isThumbnailPath(cached)
         const hasUpdate = isThumb ? (this.updateFlags.get(key) ?? false) : false
@@ -135,7 +134,6 @@ export class ImageDecryptService {
     for (const key of cacheKeys) {
       const existing = this.findCachedOutput(key, false, payload.sessionId)
       if (existing) {
-        this.logInfo('缓存命中(文件系统)', { key, path: existing, isThumb: this.isThumbnailPath(existing) })
         this.cacheResolvedPaths(key, payload.imageMd5, payload.imageDatName, existing)
         const dataUrl = this.fileToDataUrl(existing)
         const isThumb = this.isThumbnailPath(existing)
@@ -277,12 +275,12 @@ export class ImageDecryptService {
       decrypted = wxgfResult.data
 
       let ext = this.detectImageExtension(decrypted)
-      
+
       // 如果是 wxgf 格式且没检测到扩展名
       if (wxgfResult.isWxgf && !ext) {
         ext = '.hevc'
       }
-      
+
       const finalExt = ext || '.jpg'
 
       const outputPath = this.getCacheOutputPathFromDat(datPath, finalExt, payload.sessionId)
@@ -291,8 +289,8 @@ export class ImageDecryptService {
 
       // 对于 hevc 格式，返回错误提示
       if (finalExt === '.hevc') {
-        return { 
-          success: false, 
+        return {
+          success: false,
           error: '此图片为微信新格式(wxgf)，需要安装 ffmpeg 才能显示',
           isThumb: this.isThumbnailPath(datPath)
         }
@@ -1475,29 +1473,29 @@ export class ImageDecryptService {
    */
   private async unwrapWxgf(buffer: Buffer): Promise<{ data: Buffer; isWxgf: boolean }> {
     // 检查是否是 wxgf 格式 (77 78 67 66 = "wxgf")
-    if (buffer.length < 20 || 
-        buffer[0] !== 0x77 || buffer[1] !== 0x78 || 
-        buffer[2] !== 0x67 || buffer[3] !== 0x66) {
+    if (buffer.length < 20 ||
+      buffer[0] !== 0x77 || buffer[1] !== 0x78 ||
+      buffer[2] !== 0x67 || buffer[3] !== 0x66) {
       return { data: buffer, isWxgf: false }
     }
-    
+
     // 先尝试搜索内嵌的传统图片签名
     for (let i = 4; i < Math.min(buffer.length - 12, 4096); i++) {
       if (buffer[i] === 0xff && buffer[i + 1] === 0xd8 && buffer[i + 2] === 0xff) {
         return { data: buffer.subarray(i), isWxgf: false }
       }
-      if (buffer[i] === 0x89 && buffer[i + 1] === 0x50 && 
-          buffer[i + 2] === 0x4e && buffer[i + 3] === 0x47) {
+      if (buffer[i] === 0x89 && buffer[i + 1] === 0x50 &&
+        buffer[i + 2] === 0x4e && buffer[i + 3] === 0x47) {
         return { data: buffer.subarray(i), isWxgf: false }
       }
     }
-    
+
     // 提取 HEVC NALU 裸流
     const hevcData = this.extractHevcNalu(buffer)
     if (!hevcData || hevcData.length < 100) {
       return { data: buffer, isWxgf: true }
     }
-    
+
     // 尝试用 ffmpeg 转换
     try {
       const jpgData = await this.convertHevcToJpg(hevcData)
@@ -1507,7 +1505,7 @@ export class ImageDecryptService {
     } catch {
       // ffmpeg 转换失败
     }
-    
+
     return { data: hevcData, isWxgf: true }
   }
 
@@ -1517,23 +1515,23 @@ export class ImageDecryptService {
   private extractHevcNalu(buffer: Buffer): Buffer | null {
     const nalUnits: Buffer[] = []
     let i = 4
-    
+
     while (i < buffer.length - 4) {
-      if (buffer[i] === 0x00 && buffer[i + 1] === 0x00 && 
-          buffer[i + 2] === 0x00 && buffer[i + 3] === 0x01) {
+      if (buffer[i] === 0x00 && buffer[i + 1] === 0x00 &&
+        buffer[i + 2] === 0x00 && buffer[i + 3] === 0x01) {
         let nalStart = i
         let nalEnd = buffer.length
-        
+
         for (let j = i + 4; j < buffer.length - 3; j++) {
           if (buffer[j] === 0x00 && buffer[j + 1] === 0x00) {
-            if (buffer[j + 2] === 0x01 || 
-                (buffer[j + 2] === 0x00 && j + 3 < buffer.length && buffer[j + 3] === 0x01)) {
+            if (buffer[j + 2] === 0x01 ||
+              (buffer[j + 2] === 0x00 && j + 3 < buffer.length && buffer[j + 3] === 0x01)) {
               nalEnd = j
               break
             }
           }
         }
-        
+
         const nalUnit = buffer.subarray(nalStart, nalEnd)
         if (nalUnit.length > 3) {
           nalUnits.push(nalUnit)
@@ -1543,17 +1541,17 @@ export class ImageDecryptService {
         i++
       }
     }
-    
+
     if (nalUnits.length === 0) {
       for (let j = 4; j < buffer.length - 4; j++) {
-        if (buffer[j] === 0x00 && buffer[j + 1] === 0x00 && 
-            buffer[j + 2] === 0x00 && buffer[j + 3] === 0x01) {
+        if (buffer[j] === 0x00 && buffer[j + 1] === 0x00 &&
+          buffer[j + 2] === 0x00 && buffer[j + 3] === 0x01) {
           return buffer.subarray(j)
         }
       }
       return null
     }
-    
+
     return Buffer.concat(nalUnits)
   }
 
@@ -1563,11 +1561,11 @@ export class ImageDecryptService {
   private getFfmpegPath(): string {
     const staticPath = getStaticFfmpegPath()
     this.logInfo('ffmpeg 路径检测', { staticPath, exists: staticPath ? existsSync(staticPath) : false })
-    
+
     if (staticPath) {
       return staticPath
     }
-    
+
     // 回退到系统 ffmpeg
     return 'ffmpeg'
   }
@@ -1578,12 +1576,12 @@ export class ImageDecryptService {
   private convertHevcToJpg(hevcData: Buffer): Promise<Buffer | null> {
     const ffmpeg = this.getFfmpegPath()
     this.logInfo('ffmpeg 转换开始', { ffmpegPath: ffmpeg, hevcSize: hevcData.length })
-    
+
     return new Promise((resolve) => {
       const { spawn } = require('child_process')
       const chunks: Buffer[] = []
       const errChunks: Buffer[] = []
-      
+
       const proc = spawn(ffmpeg, [
         '-hide_banner',
         '-loglevel', 'error',
@@ -1593,14 +1591,14 @@ export class ImageDecryptService {
         '-q:v', '3',
         '-f', 'mjpeg',
         'pipe:1'
-      ], { 
+      ], {
         stdio: ['pipe', 'pipe', 'pipe'],
         windowsHide: true
       })
-      
+
       proc.stdout.on('data', (chunk: Buffer) => chunks.push(chunk))
       proc.stderr.on('data', (chunk: Buffer) => errChunks.push(chunk))
-      
+
       proc.on('close', (code: number) => {
         if (code === 0 && chunks.length > 0) {
           this.logInfo('ffmpeg 转换成功', { outputSize: Buffer.concat(chunks).length })
@@ -1611,12 +1609,12 @@ export class ImageDecryptService {
           resolve(null)
         }
       })
-      
+
       proc.on('error', (err: Error) => {
         this.logInfo('ffmpeg 进程错误', { error: err.message })
         resolve(null)
       })
-      
+
       proc.stdin.write(hevcData)
       proc.stdin.end()
     })

electron/services/messageCacheService.ts

@@ -15,7 +15,7 @@ export class MessageCacheService {
   constructor(cacheBasePath?: string) {
     const basePath = cacheBasePath && cacheBasePath.trim().length > 0
       ? cacheBasePath
-      : join(app.getPath('userData'), 'WeFlowCache')
+      : join(app.getPath('documents'), 'WeFlow')
     this.cacheFilePath = join(basePath, 'session-messages.json')
     this.ensureCacheDir()
     this.loadCache()

electron/services/voiceTranscribeService.ts

@@ -1,19 +1,23 @@
 import { app } from 'electron'
-import { createWriteStream, existsSync, mkdirSync, statSync, unlinkSync, writeFileSync } from 'fs'
-import { join, dirname } from 'path'
-import { promisify } from 'util'
-import { execFile, spawnSync } from 'child_process'
+import { existsSync, mkdirSync, statSync, unlinkSync, createWriteStream } from 'fs'
+import { join } from 'path'
 import * as https from 'https'
 import * as http from 'http'
 import { ConfigService } from './config'
 
-const execFileAsync = promisify(execFile)
+// Sherpa-onnx 类型定义
+type OfflineRecognizer = any
+type OfflineStream = any
 
-type WhisperModelInfo = {
+type ModelInfo = {
   name: string
-  fileName: string
+  files: {
+    model: string
+    tokens: string
+    vad: string
+  }
+  sizeBytes: number
   sizeLabel: string
-  sizeBytes?: number
 }
 
 type DownloadProgress = {
@@ -23,122 +27,169 @@ type DownloadProgress = {
   percent?: number
 }
 
-const WHISPER_MODELS: Record<string, WhisperModelInfo> = {
-  tiny: { name: 'tiny', fileName: 'ggml-tiny.bin', sizeLabel: '75 MB', sizeBytes: 75_000_000 },
-  base: { name: 'base', fileName: 'ggml-base.bin', sizeLabel: '142 MB', sizeBytes: 142_000_000 },
-  small: { name: 'small', fileName: 'ggml-small.bin', sizeLabel: '466 MB', sizeBytes: 466_000_000 },
-  medium: { name: 'medium', fileName: 'ggml-medium.bin', sizeLabel: '1.5 GB', sizeBytes: 1_500_000_000 },
-  'large-v3': { name: 'large-v3', fileName: 'ggml-large-v3.bin', sizeLabel: '2.9 GB', sizeBytes: 2_900_000_000 }
+const SENSEVOICE_MODEL: ModelInfo = {
+  name: 'SenseVoiceSmall',
+  files: {
+    model: 'model.int8.onnx',
+    tokens: 'tokens.txt',
+    vad: 'silero_vad.onnx'
+  },
+  sizeBytes: 245_000_000,
+  sizeLabel: '245 MB'
 }
 
-const WHISPER_SOURCES: Record<string, string> = {
-  official: 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main',
-  tsinghua: 'https://hf-mirror.com/ggerganov/whisper.cpp/resolve/main'
-}
-
-function getStaticFfmpegPath(): string | null {
-  try {
-    // eslint-disable-next-line @typescript-eslint/no-var-requires
-    const ffmpegStatic = require('ffmpeg-static')
-    if (typeof ffmpegStatic === 'string' && existsSync(ffmpegStatic)) {
-      return ffmpegStatic
-    }
-    const devPath = join(process.cwd(), 'node_modules', 'ffmpeg-static', 'ffmpeg.exe')
-    if (existsSync(devPath)) {
-      return devPath
-    }
-    if (app.isPackaged) {
-      const resourcesPath = process.resourcesPath
-      const packedPath = join(resourcesPath, 'app.asar.unpacked', 'node_modules', 'ffmpeg-static', 'ffmpeg.exe')
-      if (existsSync(packedPath)) {
-        return packedPath
-      }
-    }
-    return null
-  } catch {
-    return null
-  }
+const MODEL_DOWNLOAD_URLS = {
+  model: 'https://modelscope.cn/models/pengzhendong/sherpa-onnx-sense-voice-zh-en-ja-ko-yue/resolve/master/model.int8.onnx',
+  tokens: 'https://modelscope.cn/models/pengzhendong/sherpa-onnx-sense-voice-zh-en-ja-ko-yue/resolve/master/tokens.txt',
+  vad: 'https://www.modelscope.cn/models/manyeyes/silero-vad-onnx/resolve/master/silero_vad.onnx'
 }
 
 export class VoiceTranscribeService {
   private configService = new ConfigService()
   private downloadTasks = new Map<string, Promise<{ success: boolean; path?: string; error?: string }>>()
+  private recognizer: OfflineRecognizer | null = null
+  private isInitializing = false
 
-  private resolveModelInfo(modelName: string): WhisperModelInfo | null {
-    return WHISPER_MODELS[modelName] || null
-  }
-
-  private resolveModelDir(overrideDir?: string): string {
-    const configured = overrideDir || this.configService.get('whisperModelDir')
+  private resolveModelDir(): string {
+    const configured = this.configService.get('whisperModelDir') as string | undefined
     if (configured) return configured
-    return join(app.getPath('userData'), 'models', 'whisper')
+    return join(app.getPath('documents'), 'WeFlow', 'models', 'sensevoice')
   }
 
-  private resolveModelPath(modelName: string, overrideDir?: string): string | null {
-    const info = this.resolveModelInfo(modelName)
-    if (!info) return null
-    return join(this.resolveModelDir(overrideDir), info.fileName)
+  private resolveModelPath(fileName: string): string {
+    return join(this.resolveModelDir(), fileName)
   }
 
-  private resolveSourceUrl(overrideSource?: string): string {
-    const configured = overrideSource || this.configService.get('whisperDownloadSource')
-    if (configured && WHISPER_SOURCES[configured]) return WHISPER_SOURCES[configured]
-    return WHISPER_SOURCES.official
-  }
-
-  async getModelStatus(payload: { modelName: string; downloadDir?: string }): Promise<{
+  /**
+   * 检查模型状态
+   */
+  async getModelStatus(): Promise<{
     success: boolean
     exists?: boolean
-    path?: string
+    modelPath?: string
+    tokensPath?: string
     sizeBytes?: number
     error?: string
   }> {
-    const modelPath = this.resolveModelPath(payload.modelName, payload.downloadDir)
-    if (!modelPath) {
-      return { success: false, error: '未知模型名称' }
+    try {
+      const modelPath = this.resolveModelPath(SENSEVOICE_MODEL.files.model)
+      const tokensPath = this.resolveModelPath(SENSEVOICE_MODEL.files.tokens)
+      const vadPath = this.resolveModelPath((SENSEVOICE_MODEL.files as any).vad)
+
+      const modelExists = existsSync(modelPath)
+      const tokensExists = existsSync(tokensPath)
+      const vadExists = existsSync(vadPath)
+      const exists = modelExists && tokensExists && vadExists
+
+      if (!exists) {
+        return { success: true, exists: false, modelPath, tokensPath }
+      }
+
+      const modelSize = statSync(modelPath).size
+      const tokensSize = statSync(tokensPath).size
+      const vadSize = statSync(vadPath).size
+      const totalSize = modelSize + tokensSize + vadSize
+
+      return {
+        success: true,
+        exists: true,
+        modelPath,
+        tokensPath,
+        sizeBytes: totalSize
+      }
+    } catch (error) {
+      console.error('[VoiceTranscribe] getModelStatus error:', error)
+      return { success: false, error: String(error) }
     }
-    if (!existsSync(modelPath)) {
-      return { success: true, exists: false, path: modelPath }
-    }
-    const sizeBytes = statSync(modelPath).size
-    return { success: true, exists: true, path: modelPath, sizeBytes }
   }
 
+  /**
+   * 下载模型文件
+   */
   async downloadModel(
-    payload: { modelName: string; downloadDir?: string; source?: string },
     onProgress?: (progress: DownloadProgress) => void
-  ): Promise<{ success: boolean; path?: string; error?: string }> {
-    const info = this.resolveModelInfo(payload.modelName)
-    if (!info) {
-      return { success: false, error: '未知模型名称' }
-    }
-
-    const modelPath = this.resolveModelPath(payload.modelName, payload.downloadDir)
-    if (!modelPath) {
-      return { success: false, error: '模型路径生成失败' }
-    }
-
-    if (existsSync(modelPath)) {
-      return { success: true, path: modelPath }
-    }
-
-    const cacheKey = `${payload.modelName}:${modelPath}`
+  ): Promise<{ success: boolean; modelPath?: string; tokensPath?: string; error?: string }> {
+    const cacheKey = 'sensevoice'
     const pending = this.downloadTasks.get(cacheKey)
     if (pending) return pending
 
     const task = (async () => {
       try {
-        const targetDir = this.resolveModelDir(payload.downloadDir)
-        if (!existsSync(targetDir)) {
-          mkdirSync(targetDir, { recursive: true })
+        const modelDir = this.resolveModelDir()
+        if (!existsSync(modelDir)) {
+          mkdirSync(modelDir, { recursive: true })
         }
 
-        const baseUrl = this.resolveSourceUrl(payload.source)
-        const url = `${baseUrl}/${info.fileName}`
-        await this.downloadToFile(url, modelPath, payload.modelName, onProgress)
-        return { success: true, path: modelPath }
+        const modelPath = this.resolveModelPath(SENSEVOICE_MODEL.files.model)
+        const tokensPath = this.resolveModelPath(SENSEVOICE_MODEL.files.tokens)
+        const vadPath = this.resolveModelPath((SENSEVOICE_MODEL.files as any).vad)
+
+        // 下载模型文件 (40%)
+        console.info('[VoiceTranscribe] 开始下载模型文件...')
+        await this.downloadToFile(
+          MODEL_DOWNLOAD_URLS.model,
+          modelPath,
+          'model',
+          (downloaded, total) => {
+            const percent = total ? (downloaded / total) * 40 : undefined
+            onProgress?.({
+              modelName: SENSEVOICE_MODEL.name,
+              downloadedBytes: downloaded,
+              totalBytes: SENSEVOICE_MODEL.sizeBytes,
+              percent
+            })
+          }
+        )
+
+        // 下载 tokens 文件 (30%)
+        console.info('[VoiceTranscribe] 开始下载 tokens 文件...')
+        await this.downloadToFile(
+          MODEL_DOWNLOAD_URLS.tokens,
+          tokensPath,
+          'tokens',
+          (downloaded, total) => {
+            const modelSize = existsSync(modelPath) ? statSync(modelPath).size : 0
+            const percent = total ? 40 + (downloaded / total) * 30 : 40
+            onProgress?.({
+              modelName: SENSEVOICE_MODEL.name,
+              downloadedBytes: modelSize + downloaded,
+              totalBytes: SENSEVOICE_MODEL.sizeBytes,
+              percent
+            })
+          }
+        )
+
+        // 下载 vad 文件 (30%)
+        console.info('[VoiceTranscribe] 开始下载 VAD 文件...')
+        await this.downloadToFile(
+          (MODEL_DOWNLOAD_URLS as any).vad,
+          vadPath,
+          'vad',
+          (downloaded, total) => {
+            const modelSize = existsSync(modelPath) ? statSync(modelPath).size : 0
+            const tokensSize = existsSync(tokensPath) ? statSync(tokensPath).size : 0
+            const percent = total ? 70 + (downloaded / total) * 30 : 70
+            onProgress?.({
+              modelName: SENSEVOICE_MODEL.name,
+              downloadedBytes: modelSize + tokensSize + downloaded,
+              totalBytes: SENSEVOICE_MODEL.sizeBytes,
+              percent
+            })
+          }
+        )
+
+        console.info('[VoiceTranscribe] 模型下载完成')
+        return { success: true, modelPath, tokensPath }
       } catch (error) {
-        try { if (existsSync(modelPath)) unlinkSync(modelPath) } catch { }
+        console.error('[VoiceTranscribe] 下载失败:', error)
+        const modelPath = this.resolveModelPath(SENSEVOICE_MODEL.files.model)
+        const tokensPath = this.resolveModelPath(SENSEVOICE_MODEL.files.tokens)
+        const vadPath = this.resolveModelPath((SENSEVOICE_MODEL.files as any).vad)
+        try {
+          if (existsSync(modelPath)) unlinkSync(modelPath)
+          if (existsSync(tokensPath)) unlinkSync(tokensPath)
+          if (existsSync(vadPath)) unlinkSync(vadPath)
+        } catch { }
         return { success: false, error: String(error) }
       } finally {
         this.downloadTasks.delete(cacheKey)
@@ -149,102 +200,108 @@ export class VoiceTranscribeService {
     return task
   }
 
-  async transcribeWavBuffer(wavData: Buffer): Promise<{ success: boolean; transcript?: string; error?: string }> {
-    const modelName = this.configService.get('whisperModelName') || 'base'
-    const modelPath = this.resolveModelPath(modelName)
-    console.info('[VoiceTranscribe] check model', { modelName, modelPath, exists: modelPath ? existsSync(modelPath) : false })
-    if (!modelPath || !existsSync(modelPath)) {
-      return { success: false, error: '未下载语音模型，请在设置中下载' }
-    }
+  /**
+   * 转写 WAV 音频数据 (后台 Worker Threads 版本)
+   */
+  async transcribeWavBuffer(
+    wavData: Buffer,
+    onPartial?: (text: string) => void
+  ): Promise<{ success: boolean; transcript?: string; error?: string }> {
+    return new Promise((resolve) => {
+      try {
+        const modelPath = this.resolveModelPath(SENSEVOICE_MODEL.files.model)
+        const tokensPath = this.resolveModelPath(SENSEVOICE_MODEL.files.tokens)
 
-    // 使用内置的预编译 whisper-cli.exe
-    const resourcesPath = app.isPackaged
-      ? join(process.resourcesPath, 'resources')
-      : join(app.getAppPath(), 'resources')
-    const whisperExe = join(resourcesPath, 'whisper-cli.exe')
-    
-    if (!existsSync(whisperExe)) {
-      return { success: false, error: '找不到语音转写程序，请重新安装应用' }
-    }
+        if (!existsSync(modelPath) || !existsSync(tokensPath)) {
+          resolve({ success: false, error: '模型文件不存在，请先下载模型' })
+          return
+        }
 
-    const ffmpegPath = getStaticFfmpegPath() || 'ffmpeg'
-    console.info('[VoiceTranscribe] ffmpeg path', ffmpegPath)
+        const { Worker } = require('worker_threads')
+        // main.js 和 transcribeWorker.js 同在 dist-electron 目录下
+        const workerPath = join(__dirname, 'transcribeWorker.js')
 
-    const tempDir = app.getPath('temp')
-    const fileToken = `${Date.now()}_${Math.random().toString(16).slice(2)}`
-    const inputPath = join(tempDir, `weflow_voice_${fileToken}.wav`)
-    const outputPath = join(tempDir, `weflow_voice_${fileToken}_16k.wav`)
+        console.info('[VoiceTranscribe] 启动后台 Worker 转写...', { workerPath })
 
-    try {
-      writeFileSync(inputPath, wavData)
-      console.info('[VoiceTranscribe] converting to 16kHz', { inputPath, outputPath })
-      await execFileAsync(ffmpegPath, ['-y', '-i', inputPath, '-ar', '16000', '-ac', '1', outputPath])
-      
-      console.info('[VoiceTranscribe] transcribing with whisper', { whisperExe, modelPath })
-      const { stdout, stderr } = await execFileAsync(whisperExe, [
-        '-m', modelPath,
-        '-f', outputPath,
-        '-l', 'zh',
-        '-otxt',
-        '-np'  // no prints (只输出结果)
-      ], {
-        maxBuffer: 10 * 1024 * 1024,
-        cwd: dirname(whisperExe),  // 设置工作目录为 whisper-cli.exe 所在目录，确保能找到 DLL
-        env: { ...process.env, PATH: `${dirname(whisperExe)};${process.env.PATH}` }
-      })
+        const worker = new Worker(workerPath, {
+          workerData: {
+            modelPath,
+            tokensPath,
+            wavData,
+            sampleRate: 16000
+          }
+        })
 
-      console.info('[VoiceTranscribe] whisper stdout:', stdout)
-      if (stderr) console.warn('[VoiceTranscribe] whisper stderr:', stderr)
+        let finalTranscript = ''
 
-      // 解析输出文本
-      const outputBase = outputPath.replace(/\.[^.]+$/, '')
-      const txtFile = `${outputBase}.txt`
-      let transcript = ''
-      if (existsSync(txtFile)) {
-        const { readFileSync } = await import('fs')
-        transcript = readFileSync(txtFile, 'utf-8').trim()
-        unlinkSync(txtFile)
-      } else {
-        // 从 stdout 提取（使用 -np 参数后，stdout 只有转写结果）
-        transcript = stdout.trim()
+        worker.on('message', (msg: any) => {
+          if (msg.type === 'partial') {
+            onPartial?.(msg.text)
+          } else if (msg.type === 'final') {
+            finalTranscript = msg.text
+            resolve({ success: true, transcript: finalTranscript })
+            worker.terminate()
+          } else if (msg.type === 'error') {
+            resolve({ success: false, error: msg.error })
+            worker.terminate()
+          }
+        })
+
+        worker.on('error', (err: Error) => {
+          console.error('[VoiceTranscribe] Worker error:', err)
+          resolve({ success: false, error: String(err) })
+        })
+
+        worker.on('exit', (code: number) => {
+          if (code !== 0) {
+            console.error(`[VoiceTranscribe] Worker stopped with exit code ${code}`)
+            resolve({ success: false, error: `Worker exited with code ${code}` })
+          }
+        })
+
+      } catch (error) {
+        console.error('[VoiceTranscribe] 启动 Worker 失败:', error)
+        resolve({ success: false, error: String(error) })
       }
-
-      console.info('[VoiceTranscribe] success', { transcript })
-      return { success: true, transcript }
-    } catch (error: any) {
-      console.error('[VoiceTranscribe] failed', error)
-      console.error('[VoiceTranscribe] stderr:', error.stderr)
-      console.error('[VoiceTranscribe] stdout:', error.stdout)
-      return { success: false, error: String(error) }
-    } finally {
-      try { if (existsSync(inputPath)) unlinkSync(inputPath) } catch { }
-      try { if (existsSync(outputPath)) unlinkSync(outputPath) } catch { }
-    }
+    })
   }
 
+  /**
+   * 下载文件
+   */
   private downloadToFile(
     url: string,
     targetPath: string,
-    modelName: string,
-    onProgress?: (progress: DownloadProgress) => void,
-    remainingRedirects = 3
+    fileName: string,
+    onProgress?: (downloaded: number, total?: number) => void,
+    remainingRedirects = 5
   ): Promise<void> {
     return new Promise((resolve, reject) => {
       const protocol = url.startsWith('https') ? https : http
-      const request = protocol.get(url, (response) => {
+      console.info(`[VoiceTranscribe] 下载 ${fileName}:`, url)
+
+      const options = {
+        headers: {
+          'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36'
+        }
+      }
+
+      const request = protocol.get(url, options, (response) => {
+        // 处理重定向
         if ([301, 302, 303, 307, 308].includes(response.statusCode || 0) && response.headers.location) {
           if (remainingRedirects <= 0) {
-            reject(new Error('下载重定向次数过多'))
+            reject(new Error('重定向次数过多'))
             return
           }
-          this.downloadToFile(response.headers.location, targetPath, modelName, onProgress, remainingRedirects - 1)
+          console.info(`[VoiceTranscribe] 重定向到:`, response.headers.location)
+          this.downloadToFile(response.headers.location, targetPath, fileName, onProgress, remainingRedirects - 1)
             .then(resolve)
             .catch(reject)
           return
         }
 
         if (response.statusCode !== 200) {
-          reject(new Error(`下载失败: ${response.statusCode}`))
+          reject(new Error(`下载失败: HTTP ${response.statusCode}`))
           return
         }
 
@@ -255,8 +312,7 @@ export class VoiceTranscribeService {
 
         response.on('data', (chunk) => {
           downloadedBytes += chunk.length
-          const percent = totalBytes ? (downloadedBytes / totalBytes) * 100 : undefined
-          onProgress?.({ modelName, downloadedBytes, totalBytes, percent })
+          onProgress?.(downloadedBytes, totalBytes)
         })
 
         response.on('error', (error) => {
@@ -271,15 +327,33 @@ export class VoiceTranscribeService {
 
         writer.on('finish', () => {
           writer.close()
+          console.info(`[VoiceTranscribe] ${fileName} 下载完成:`, targetPath)
           resolve()
         })
 
         response.pipe(writer)
       })
 
-      request.on('error', reject)
+      request.on('error', (error) => {
+        console.error(`[VoiceTranscribe] ${fileName} 下载错误:`, error)
+        reject(error)
+      })
     })
   }
+
+  /**
+   * 清理资源
+   */
+  dispose() {
+    if (this.recognizer) {
+      try {
+        // sherpa-onnx 的 recognizer 可能需要手动释放
+        this.recognizer = null
+      } catch (error) {
+        console.error('[VoiceTranscribe] 释放识别器失败:', error)
+      }
+    }
+  }
 }
 
 export const voiceTranscribeService = new VoiceTranscribeService()

electron/services/wcdbCore.ts

@@ -48,6 +48,7 @@ export class WcdbCore {
   private wcdbGetMessageById: any = null
   private wcdbGetEmoticonCdnUrl: any = null
   private wcdbGetDbStatus: any = null
+  private wcdbGetVoiceData: any = null
   private avatarUrlCache: Map<string, { url?: string; updatedAt: number }> = new Map()
   private readonly avatarCacheTtlMs = 10 * 60 * 1000
   private logTimer: NodeJS.Timeout | null = null
@@ -108,12 +109,13 @@ export class WcdbCore {
 
   private writeLog(message: string, force = false): void {
     if (!force && !this.isLogEnabled()) return
+    const line = `[${new Date().toISOString()}] ${message}`
+    console.log(`[WCDB] ${line}`)
     try {
       const base = this.userDataPath || process.env.WCDB_LOG_DIR || process.cwd()
       const dir = join(base, 'logs')
       if (!existsSync(dir)) mkdirSync(dir, { recursive: true })
-      const line = `[${new Date().toISOString()}] ${message}\n`
-      appendFileSync(join(dir, 'wcdb.log'), line, { encoding: 'utf8' })
+      appendFileSync(join(dir, 'wcdb.log'), line + '\n', { encoding: 'utf8' })
     } catch { }
   }
 
@@ -345,6 +347,13 @@ export class WcdbCore {
         this.wcdbGetDbStatus = null
       }
 
+      // wcdb_status wcdb_get_voice_data(wcdb_handle handle, const char* session_id, int32_t create_time, const char* candidates_json, char** out_hex)
+      try {
+        this.wcdbGetVoiceData = this.lib.func('int32 wcdb_get_voice_data(int64 handle, const char* sessionId, int32 createTime, int64 svrId, const char* candidatesJson, _Out_ void** outHex)')
+      } catch {
+        this.wcdbGetVoiceData = null
+      }
+
       // 初始化
       const initResult = this.wcdbInit()
       if (initResult !== 0) {
@@ -1295,9 +1304,7 @@ export class WcdbCore {
     } catch (e) {
       return { success: false, error: String(e) }
     }
-  }
-
-  async getMessageById(sessionId: string, localId: number): Promise<{ success: boolean; message?: any; error?: string }> {
+  } async getMessageById(sessionId: string, localId: number): Promise<{ success: boolean; message?: any; error?: string }> {
     if (!this.ensureReady()) return { success: false, error: 'WCDB 未连接' }
     try {
       const outPtr = [null as any]
@@ -1313,5 +1320,21 @@ export class WcdbCore {
       return { success: false, error: String(e) }
     }
   }
-}
 
+  async getVoiceData(sessionId: string, createTime: number, candidates: string[], svrId: string | number = 0): Promise<{ success: boolean; hex?: string; error?: string }> {
+    if (!this.ensureReady()) return { success: false, error: 'WCDB 未连接' }
+    if (!this.wcdbGetVoiceData) return { success: false, error: '当前 DLL 版本不支持获取语音数据' }
+    try {
+      const outPtr = [null as any]
+      const result = this.wcdbGetVoiceData(this.handle, sessionId, createTime, BigInt(svrId || 0), JSON.stringify(candidates), outPtr)
+      if (result !== 0 || !outPtr[0]) {
+        return { success: false, error: `获取语音数据失败: ${result}` }
+      }
+      const hex = this.decodeJsonPtr(outPtr[0])
+      if (hex === null) return { success: false, error: '解析语音数据失败' }
+      return { success: true, hex: hex || undefined }
+    } catch (e) {
+      return { success: false, error: String(e) }
+    }
+  }
+}

electron/services/wcdbService.ts

@@ -341,6 +341,13 @@ export class WcdbService {
     return this.callWorker('getMessageById', { sessionId, localId })
   }
 
+  /**
+   * 获取语音数据
+   */
+  async getVoiceData(sessionId: string, createTime: number, candidates: string[], svrId: string | number = 0): Promise<{ success: boolean; hex?: string; error?: string }> {
+    return this.callWorker('getVoiceData', { sessionId, createTime, candidates, svrId })
+  }
+
 }
 
 export const wcdbService = new WcdbService()

electron/transcribeWorker.ts

@@ -0,0 +1,174 @@
+import { parentPort, workerData } from 'worker_threads'
+import * as fs from 'fs'
+
+interface WorkerParams {
+    modelPath: string
+    tokensPath: string
+    wavData: Buffer
+    sampleRate: number
+}
+
+async function run() {
+    console.info('[TranscribeWorker] Worker process starting...');
+
+    if (!parentPort) {
+        console.error('[TranscribeWorker] Critical Error: parentPort is null');
+        return;
+    }
+
+    try {
+        console.info('[TranscribeWorker] Loading sherpa-onnx-node...');
+        // 动态加载以捕获可能的加载错误（如 C++ 运行库缺失等）
+        let sherpa: any;
+        try {
+            sherpa = require('sherpa-onnx-node');
+            console.info('[TranscribeWorker] sherpa-onnx-node loaded successfully.');
+        } catch (requireError) {
+            console.error('[TranscribeWorker] Failed to load sherpa-onnx-node:', requireError);
+            parentPort.postMessage({ type: 'error', error: 'Failed to load speech engine: ' + String(requireError) });
+            return;
+        }
+
+        const { modelPath, tokensPath, wavData: rawWavData, sampleRate } = workerData as WorkerParams
+        const wavData = Buffer.from(rawWavData);
+        console.info('[TranscribeWorker] Params received:', {
+            modelPath,
+            tokensPath,
+            sampleRate,
+            wavDataLength: wavData?.length
+        });
+
+        // 1. 初始化识别器 (SenseVoiceSmall)
+        console.info('[TranscribeWorker] Initializing OfflineRecognizer...');
+        const recognizerConfig = {
+            modelConfig: {
+                senseVoice: {
+                    model: modelPath,
+                    useInverseTextNormalization: 1
+                },
+                tokens: tokensPath,
+                numThreads: 2,
+                debug: 0
+            }
+        }
+        const recognizer = new sherpa.OfflineRecognizer(recognizerConfig)
+        console.info('[TranscribeWorker] OfflineRecognizer initialized.');
+
+        // 2. 初始化 VAD (用于流式输出效果)
+        const vadPath = modelPath.replace('model.int8.onnx', 'silero_vad.onnx');
+        console.info('[TranscribeWorker] VAD Path:', vadPath);
+
+        const vadConfig = {
+            sileroVad: {
+                model: vadPath,
+                threshold: 0.5,
+                minSilenceDuration: 0.5,
+                minSpeechDuration: 0.25,
+                windowSize: 512
+            },
+            sampleRate: sampleRate,
+            debug: 0,
+            numThreads: 1
+        }
+
+        // 检查 VAD 模型是否存在，如果不存在则退回到全量识别
+        if (!fs.existsSync(vadPath)) {
+            console.warn('[TranscribeWorker] VAD model not found, falling back to full transcription.');
+
+            const pcmData = wavData.slice(44)
+            const samples = new Float32Array(pcmData.length / 2)
+            for (let i = 0; i < samples.length; i++) {
+                samples[i] = pcmData.readInt16LE(i * 2) / 32768.0
+            }
+
+            const stream = recognizer.createStream()
+            stream.acceptWaveform({ sampleRate, samples })
+            recognizer.decode(stream)
+            const result = recognizer.getResult(stream)
+
+            console.info('[TranscribeWorker] Full transcription result:', result.text);
+            parentPort.postMessage({ type: 'final', text: result.text })
+            return
+        }
+
+        console.info('[TranscribeWorker] Initializing Vad...');
+        const vad = new sherpa.Vad(vadConfig, 60) // 60s max
+        console.info('[TranscribeWorker] VAD initialized.');
+
+        // 3. 处理音频数据
+        const pcmData = wavData.slice(44)
+        const samples = new Float32Array(pcmData.length / 2)
+        for (let i = 0; i < samples.length; i++) {
+            samples[i] = pcmData.readInt16LE(i * 2) / 32768.0
+        }
+
+        // 模拟流式输入：按小块喂给 VAD
+        const chunkSize = 1600 // 100ms for 16kHz
+        let offset = 0
+        let accumulatedText = ''
+
+        console.info('[TranscribeWorker] Starting processing loop...');
+        let segmentCount = 0;
+
+        while (offset < samples.length) {
+            const end = Math.min(offset + chunkSize, samples.length)
+            const chunk = samples.subarray(offset, end)
+
+            vad.acceptWaveform(chunk)
+
+            // 检查 ASR 结果
+            while (!vad.isEmpty()) {
+                const segment = vad.front(false)
+
+                // Log segment detection
+                console.info(`[TranscribeWorker] VAD Segment detected. Duration: ${segment.samples.length / sampleRate}s`);
+
+                const stream = recognizer.createStream()
+                stream.acceptWaveform({ sampleRate, samples: segment.samples })
+                recognizer.decode(stream)
+                const result = recognizer.getResult(stream)
+
+                if (result.text) {
+                    const text = result.text.trim();
+                    if (text.length > 0) {
+                        accumulatedText += (accumulatedText ? ' ' : '') + text
+                        segmentCount++;
+                        console.info(`[TranscribeWorker] Partial update #${segmentCount}: "${text}" -> Total: "${accumulatedText.substring(0, 50)}..."`);
+                        parentPort.postMessage({ type: 'partial', text: accumulatedText })
+                    }
+                }
+                vad.pop()
+            }
+
+            offset = end
+            // 让出主循环，保持响应
+            await new Promise(resolve => setImmediate(resolve))
+        }
+
+        // Ensure any remaining buffer is processed
+        vad.flush();
+        while (!vad.isEmpty()) {
+            const segment = vad.front(false);
+            console.info(`[TranscribeWorker] Final VAD Segment detected. Duration: ${segment.samples.length / sampleRate}s`);
+            const stream = recognizer.createStream()
+            stream.acceptWaveform({ sampleRate, samples: segment.samples })
+            recognizer.decode(stream)
+            const result = recognizer.getResult(stream)
+            if (result.text) {
+                accumulatedText += (accumulatedText ? ' ' : '') + result.text.trim()
+                console.info(`[TranscribeWorker] Final partial update: "${result.text.trim()}"`);
+                parentPort.postMessage({ type: 'partial', text: accumulatedText })
+            }
+            vad.pop();
+        }
+
+        console.info('[TranscribeWorker] Loop finished. Final text length:', accumulatedText.length);
+        parentPort.postMessage({ type: 'final', text: accumulatedText })
+
+    } catch (error) {
+        console.error('[TranscribeWorker] Fatal error:', error);
+        parentPort.postMessage({ type: 'error', error: String(error) })
+    }
+}
+
+run();

electron/types/sherpa-onnx-node.d.ts

@@ -0,0 +1,4 @@
+declare module 'sherpa-onnx-node' {
+    const content: any;
+    export = content;
+}

electron/wcdbWorker.ts

@@ -110,6 +110,12 @@ if (parentPort) {
                 case 'getMessageById':
                     result = await core.getMessageById(payload.sessionId, payload.localId)
                     break
+                case 'getVoiceData':
+                    result = await core.getVoiceData(payload.sessionId, payload.createTime, payload.candidates, payload.svrId)
+                    if (!result.success) {
+                        console.error('[wcdbWorker] getVoiceData failed:', result.error)
+                    }
+                    break
                 default:
                     result = { success: false, error: `Unknown method: ${type}` }
             }