feat(voice-transcribe): 新增语音转写语言过滤配置功能（支持用户自定义允许的转写语言），优化模型下载的超时处理与进度日志，提升下载稳健性，同步更新相关 UI 样式。

2026-03-24 23:06:51 +00:00 · 2026-01-17 19:54:31 +08:00
parent dc12df0fcf
commit 0853e049c8
9 changed files with 338 additions and 66 deletions
--- a/electron/services/config.ts
+++ b/electron/services/config.ts
@@ -23,6 +23,8 @@ interface ConfigSchema {
  whisperModelName: string
  whisperModelDir: string
  whisperDownloadSource: string
+  autoTranscribeVoice: boolean
+  transcribeLanguages: string[]
 }

 export class ConfigService {
@@ -48,7 +50,9 @@ export class ConfigService {
        llmModelPath: '',
        whisperModelName: 'base',
        whisperModelDir: '',
-        whisperDownloadSource: 'tsinghua'
+        whisperDownloadSource: 'tsinghua',
+        autoTranscribeVoice: false,
+        transcribeLanguages: ['zh']
      }
    })
  }
--- a/electron/services/voiceTranscribeService.ts
+++ b/electron/services/voiceTranscribeService.ts
@@ -123,7 +123,16 @@ export class VoiceTranscribeService {
        const tokensPath = this.resolveModelPath(SENSEVOICE_MODEL.files.tokens)
        const vadPath = this.resolveModelPath((SENSEVOICE_MODEL.files as any).vad)

+        // 初始进度
+        onProgress?.({
+          modelName: SENSEVOICE_MODEL.name,
+          downloadedBytes: 0,
+          totalBytes: SENSEVOICE_MODEL.sizeBytes,
+          percent: 0
+        })
+
        // 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,
@@ -140,6 +149,7 @@ export class VoiceTranscribeService {
        )

        // 下载 tokens 文件 (30%)
+        console.info('[VoiceTranscribe] 开始下载 tokens 文件...')
        await this.downloadToFile(
          MODEL_DOWNLOAD_URLS.tokens,
          tokensPath,
@@ -157,6 +167,7 @@ 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,
@@ -174,6 +185,7 @@ export class VoiceTranscribeService {
          }
        )

+        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)
@@ -199,7 +211,8 @@ export class VoiceTranscribeService {
   */
  async transcribeWavBuffer(
    wavData: Buffer,
-    onPartial?: (text: string) => void
+    onPartial?: (text: string) => void,
+    languages?: string[]
  ): Promise<{ success: boolean; transcript?: string; error?: string }> {
    return new Promise((resolve) => {
      try {
@@ -211,6 +224,16 @@ export class VoiceTranscribeService {
          return
        }

+        // 获取配置的语言列表，如果没有传入则从配置读取
+        let supportedLanguages = languages
+        if (!supportedLanguages || supportedLanguages.length === 0) {
+          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']
+          }
+        }
+
        const { Worker } = require('worker_threads')
        // main.js 和 transcribeWorker.js 同在 dist-electron 目录下
        const workerPath = join(__dirname, 'transcribeWorker.js')
@@ -220,7 +243,8 @@ export class VoiceTranscribeService {
            modelPath,
            tokensPath,
            wavData,
-            sampleRate: 16000
+            sampleRate: 16000,
+            languages: supportedLanguages
          }
        })

@@ -273,10 +297,13 @@ export class VoiceTranscribeService {
      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'
-        }
+        },
+        timeout: 30000 // 30秒连接超时
      }

      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) {
@@ -297,25 +324,44 @@ 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(() => {
+          if (Date.now() - lastDataTime > 60000) {
+            clearInterval(dataTimeout)
+            response.destroy()
+            writer.close()
+            reject(new Error('下载超时：60秒内未收到数据'))
+          }
+        }, 5000)

        response.on('data', (chunk) => {
+          lastDataTime = Date.now()
          downloadedBytes += chunk.length
          onProgress?.(downloadedBytes, totalBytes)
        })

        response.on('error', (error) => {
+          clearInterval(dataTimeout)
          try { writer.close() } catch { }
+          console.error(`[VoiceTranscribe] ${fileName} 响应错误:`, error)
          reject(error)
        })

        writer.on('error', (error) => {
+          clearInterval(dataTimeout)
          try { writer.close() } catch { }
+          console.error(`[VoiceTranscribe] ${fileName} 写入错误:`, error)
          reject(error)
        })

        writer.on('finish', () => {
+          clearInterval(dataTimeout)
          writer.close()
          console.info(`[VoiceTranscribe] ${fileName} 下载完成:`, targetPath)
          resolve()
@@ -324,8 +370,14 @@ export class VoiceTranscribeService {
        response.pipe(writer)
      })

+      request.on('timeout', () => {
+        request.destroy()
+        console.error(`[VoiceTranscribe] ${fileName} 连接超时`)
+        reject(new Error('连接超时'))
+      })
+
      request.on('error', (error) => {
-        console.error(`[VoiceTranscribe] ${fileName} 下载错误:`, error)
+        console.error(`[VoiceTranscribe] ${fileName} 请求错误:`, error)
        reject(error)
      })
    })
--- a/electron/transcribeWorker.ts
+++ b/electron/transcribeWorker.ts
@@ -6,6 +6,43 @@ interface WorkerParams {
    tokensPath: string
    wavData: Buffer
    sampleRate: number
+    languages?: string[]
+}
+
+// 语言标记映射
+const LANGUAGE_TAGS: Record<string, string> = {
+    'zh': '<|zh|>',
+    'en': '<|en|>',
+    'ja': '<|ja|>',
+    'ko': '<|ko|>',
+    'yue': '<|yue|>' // 粤语
+}
+
+// 检查识别结果是否在允许的语言列表中
+function isLanguageAllowed(result: any, allowedLanguages: string[]): boolean {
+    if (!result || !result.lang) {
+        // 如果没有语言信息，默认允许
+        return true
+    }
+
+    // 如果没有指定语言或语言列表为空，默认只允许中文
+    if (!allowedLanguages || allowedLanguages.length === 0) {
+        allowedLanguages = ['zh']
+    }
+
+    const langTag = result.lang
+    console.log('[TranscribeWorker] 检测到语言标记:', langTag)
+
+    // 检查是否在允许的语言列表中
+    for (const lang of allowedLanguages) {
+        if (LANGUAGE_TAGS[lang] === langTag) {
+            console.log('[TranscribeWorker] 语言匹配，允许:', lang)
+            return true
+        }
+    }
+
+    console.log('[TranscribeWorker] 语言不在白名单中，过滤掉')
+    return false
 }

 async function run() {
@@ -23,8 +60,16 @@ async function run() {
            return;
        }

-        const { modelPath, tokensPath, wavData: rawWavData, sampleRate } = workerData as WorkerParams
+        const { modelPath, tokensPath, wavData: rawWavData, sampleRate, languages } = workerData as WorkerParams
        const wavData = Buffer.from(rawWavData);
+        // 确保有有效的语言列表，默认只允许中文
+        let allowedLanguages = languages || ['zh']
+        if (allowedLanguages.length === 0) {
+          allowedLanguages = ['zh']
+        }
+        
+        console.log('[TranscribeWorker] 使用的语言白名单:', allowedLanguages)
+        
        // 1. 初始化识别器 (SenseVoiceSmall)
        const recognizerConfig = {
            modelConfig: {
@@ -66,7 +111,16 @@ async function run() {
            recognizer.decode(stream)
            const result = recognizer.getResult(stream)

-            parentPort.postMessage({ type: 'final', text: result.text })
+            console.log('[TranscribeWorker] 非VAD模式 - 识别结果对象:', JSON.stringify(result, null, 2))
+            
+            // 检查语言是否在白名单中
+            if (isLanguageAllowed(result, allowedLanguages)) {
+                console.log('[TranscribeWorker] 非VAD模式 - 保留文本:', result.text)
+                parentPort.postMessage({ type: 'final', text: result.text })
+            } else {
+                console.log('[TranscribeWorker] 非VAD模式 - 语言不匹配，返回空文本')
+                parentPort.postMessage({ type: 'final', text: '' })
+            }
            return
        }

@@ -100,13 +154,18 @@ async function run() {
                recognizer.decode(stream)
                const result = recognizer.getResult(stream)

-                if (result.text) {
-                    const text = result.text.trim();
+                console.log('[TranscribeWorker] 识别结果 - lang:', result.lang, 'text:', result.text)
+                
+                // 检查语言是否在白名单中
+                if (result.text && isLanguageAllowed(result, allowedLanguages)) {
+                    const text = result.text.trim()
                    if (text.length > 0) {
                        accumulatedText += (accumulatedText ? ' ' : '') + text
                        segmentCount++;
                        parentPort.postMessage({ type: 'partial', text: accumulatedText })
                    }
+                } else if (result.text) {
+                    console.log('[TranscribeWorker] 跳过不匹配的语言段落')
                }
                vad.pop()
            }
@@ -124,9 +183,16 @@ async function run() {
            stream.acceptWaveform({ sampleRate, samples: segment.samples })
            recognizer.decode(stream)
            const result = recognizer.getResult(stream)
-            if (result.text) {
-                accumulatedText += (accumulatedText ? ' ' : '') + result.text.trim()
-                parentPort.postMessage({ type: 'partial', text: accumulatedText })
+            
+            console.log('[TranscribeWorker] flush阶段 - lang:', result.lang, 'text:', result.text)
+            
+            // 检查语言是否在白名单中
+            if (result.text && isLanguageAllowed(result, allowedLanguages)) {
+                const text = result.text.trim()
+                if (text) {
+                    accumulatedText += (accumulatedText ? ' ' : '') + text
+                    parentPort.postMessage({ type: 'partial', text: accumulatedText })
+                }
            }
            vad.pop();
        }