feat: 优化了语音配置页面的效果；新增语音实际波形图显示；新增语音点击跳转进度

fix: 修复了一个可能导致语音解密错乱的问题
2026-03-25 07:16:51 +00:00 · 2026-01-18 00:01:07 +08:00
parent 0853e049c8
commit be4d9b510d
16 changed files with 567 additions and 291 deletions
--- a/electron/services/chatService.ts
+++ b/electron/services/chatService.ts
@@ -2202,7 +2202,7 @@ class ChatService {


      // 3. 调用 C++ 接口获取语音 (Hex)
-      const voiceRes = await wcdbService.getVoiceData(sessionId, msgCreateTime, candidates, msgSvrId)
+      const voiceRes = await wcdbService.getVoiceData(sessionId, msgCreateTime, candidates, localId, msgSvrId)
      if (!voiceRes.success || !voiceRes.hex) {
        return { success: false, error: voiceRes.error || '未找到语音数据' }
      }
@@ -2245,6 +2245,33 @@ class ChatService {
    }
  }

+  /**
+   * 检查语音是否已有缓存
+   */
+  async resolveVoiceCache(sessionId: string, msgId: string): Promise<{ success: boolean; hasCache: boolean; data?: string }> {
+    try {
+      const cacheKey = this.getVoiceCacheKey(sessionId, msgId)
+
+      // 1. 检查内存缓存
+      const inMemory = this.voiceWavCache.get(cacheKey)
+      if (inMemory) {
+        return { success: true, hasCache: true, data: inMemory.toString('base64') }
+      }
+
+      // 2. 检查文件缓存
+      const cachedFile = this.getVoiceCacheFilePath(cacheKey)
+      if (existsSync(cachedFile)) {
+        const wavData = readFileSync(cachedFile)
+        this.cacheVoiceWav(cacheKey, wavData) // 回甜内存
+        return { success: true, hasCache: true, data: wavData.toString('base64') }
+      }
+
+      return { success: true, hasCache: false }
+    } catch (e) {
+      return { success: false, hasCache: false }
+    }
+  }
+
  async getVoiceData_Legacy(sessionId: string, msgId: string): Promise<{ success: boolean; data?: string; error?: string }> {
    try {
      const localId = parseInt(msgId, 10)
--- a/electron/services/voiceTranscribeService.ts
+++ b/electron/services/voiceTranscribeService.ts
@@ -14,7 +14,6 @@ type ModelInfo = {
  files: {
    model: string
    tokens: string
-    vad: string
  }
  sizeBytes: number
  sizeLabel: string
@@ -31,8 +30,7 @@ const SENSEVOICE_MODEL: ModelInfo = {
  name: 'SenseVoiceSmall',
  files: {
    model: 'model.int8.onnx',
-    tokens: 'tokens.txt',
-    vad: 'silero_vad.onnx'
+    tokens: 'tokens.txt'
  },
  sizeBytes: 245_000_000,
  sizeLabel: '245 MB'
@@ -40,8 +38,7 @@ const SENSEVOICE_MODEL: ModelInfo = {

 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'
+  tokens: 'https://modelscope.cn/models/pengzhendong/sherpa-onnx-sense-voice-zh-en-ja-ko-yue/resolve/master/tokens.txt'
 }

 export class VoiceTranscribeService {
@@ -74,12 +71,9 @@ export class VoiceTranscribeService {
    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
+      const exists = modelExists && tokensExists

      if (!exists) {
        return { success: true, exists: false, modelPath, tokensPath }
@@ -87,8 +81,7 @@ export class VoiceTranscribeService {

      const modelSize = statSync(modelPath).size
      const tokensSize = statSync(tokensPath).size
-      const vadSize = statSync(vadPath).size
-      const totalSize = modelSize + tokensSize + vadSize
+      const totalSize = modelSize + tokensSize

      return {
        success: true,
@@ -121,7 +114,6 @@ export class VoiceTranscribeService {

        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)

        // 初始进度
        onProgress?.({
@@ -166,35 +158,16 @@ export class VoiceTranscribeService {
          }
        )

-        // 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')

        console.info('[VoiceTranscribe] 所有文件下载完成')
        return { success: true, modelPath, tokensPath }
      } catch (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 {
@@ -230,7 +203,7 @@ export class VoiceTranscribeService {
          supportedLanguages = this.configService.get('transcribeLanguages')
          // 如果配置中也没有或为空，使用默认值
          if (!supportedLanguages || supportedLanguages.length === 0) {
-            supportedLanguages = ['zh']
+            supportedLanguages = ['zh', 'yue']
          }
        }

@@ -303,7 +276,7 @@ export class VoiceTranscribeService {

      const request = protocol.get(url, options, (response) => {
        console.info(`[VoiceTranscribe] ${fileName} 响应状态:`, response.statusCode)
-        
+
        // 处理重定向
        if ([301, 302, 303, 307, 308].includes(response.statusCode || 0) && response.headers.location) {
          if (remainingRedirects <= 0) {
@@ -324,11 +297,11 @@ export class VoiceTranscribeService {

        const totalBytes = Number(response.headers['content-length'] || 0) || undefined
        let downloadedBytes = 0
-        
+
        console.info(`[VoiceTranscribe] ${fileName} 文件大小:`, totalBytes ? `${(totalBytes / 1024 / 1024).toFixed(2)} MB` : '未知')

        const writer = createWriteStream(targetPath)
-        
+
        // 设置数据接收超时（60秒没有数据则超时）
        let lastDataTime = Date.now()
        const dataTimeout = setInterval(() => {
@@ -392,7 +365,7 @@ export class VoiceTranscribeService {
        // sherpa-onnx 的 recognizer 可能需要手动释放
        this.recognizer = null
      } catch (error) {
-        }
+      }
    }
  }
 }
--- a/electron/services/wcdbCore.ts
+++ b/electron/services/wcdbCore.ts
@@ -347,9 +347,9 @@ 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)
+      // wcdb_status wcdb_get_voice_data(wcdb_handle handle, const char* session_id, int32_t create_time, int32_t local_id, int64_t svr_id, 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)')
+        this.wcdbGetVoiceData = this.lib.func('int32 wcdb_get_voice_data(int64 handle, const char* sessionId, int32 createTime, int32 localId, int64 svrId, const char* candidatesJson, _Out_ void** outHex)')
      } catch {
        this.wcdbGetVoiceData = null
      }
@@ -1321,12 +1321,12 @@ export class WcdbCore {
    }
  }

-  async getVoiceData(sessionId: string, createTime: number, candidates: string[], svrId: string | number = 0): Promise<{ success: boolean; hex?: string; error?: string }> {
+  async getVoiceData(sessionId: string, createTime: number, candidates: string[], localId: number = 0, 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)
+      const result = this.wcdbGetVoiceData(this.handle, sessionId, createTime, localId, 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}` }
      }
--- a/electron/services/wcdbService.ts
+++ b/electron/services/wcdbService.ts
@@ -99,7 +99,7 @@ export class WcdbService {
  setPaths(resourcesPath: string, userDataPath: string): void {
    this.resourcesPath = resourcesPath
    this.userDataPath = userDataPath
-    this.callWorker('setPaths', { resourcesPath, userDataPath }).catch(() => {})
+    this.callWorker('setPaths', { resourcesPath, userDataPath }).catch(() => { })
  }

  /**
@@ -107,7 +107,7 @@ export class WcdbService {
   */
  setLogEnabled(enabled: boolean): void {
    this.logEnabled = enabled
-    this.callWorker('setLogEnabled', { enabled }).catch(() => {})
+    this.callWorker('setLogEnabled', { enabled }).catch(() => { })
  }

  /**
@@ -346,8 +346,8 @@ export class WcdbService {
  /**
   * 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 })
+  async getVoiceData(sessionId: string, createTime: number, candidates: string[], localId: number = 0, svrId: string | number = 0): Promise<{ success: boolean; hex?: string; error?: string }> {
+    return this.callWorker('getVoiceData', { sessionId, createTime, candidates, localId, svrId })
  }

 }