@hsiehchenwei/mcp-gemini-transcriber 1.0.0 → 1.1.0

package/package.json

@@ -1,11 +1,11 @@
 {
   "name": "@hsiehchenwei/mcp-gemini-transcriber",
-  "version": "1.0.0",
+  "version": "1.1.0",
   "type": "module",
   "description": "MCP 音訊轉逐字稿工具（使用 Gemini API）- 支援語者識別與情緒分析",
   "main": "server.mjs",
   "bin": {
-    "mcp-gemini-transcriber": "./server.mjs"
+    "mcp-gemini-transcriber": "server.mjs"
   },
   "scripts": {
     "start": "node server.mjs"
@@ -24,7 +24,7 @@
   "license": "MIT",
   "repository": {
     "type": "git",
-    "url": "https://github.com/chenwei/MCPTools.git",
+    "url": "git+https://github.com/chenwei/MCPTools.git",
     "directory": "mcp-gemini-transcriber"
   },
   "dependencies": {

package/server.mjs

@@ -45,6 +45,138 @@ const MODEL_NAME = 'gemini-3-flash-preview';
 const SUPPORTED_AUDIO_FORMATS = ['.mp3', '.m4a', '.wav', '.webm', '.ogg', '.flac', '.aiff', '.aac'];
 const SUPPORTED_IMAGE_FORMATS = ['.png', '.jpg', '.jpeg', '.webp', '.heic', '.heif'];
 
+// Gemini 3 Flash Preview 價格（每 100 萬 tokens，美元）
+const PRICING = {
+  input: {
+    text: 0.50,    // 文字/圖片/影片
+    audio: 1.00    // 音訊
+  },
+  output: 3.00     // 輸出（含思考代幣）
+};
+
+/**
+ * Token 用量追蹤器
+ */
+class UsageTracker {
+  constructor() {
+    this.reset();
+  }
+  
+  reset() {
+    this.inputTokens = { text: 0, audio: 0, image: 0, video: 0 };
+    this.outputTokens = 0;
+    this.thoughtTokens = 0;
+    this.apiCalls = 0;
+  }
+  
+  // 從 API response 中提取並累加 usage
+  addFromResponse(response) {
+    const usage = response?.usageMetadata;
+    if (!usage) return;
+    
+    this.apiCalls++;
+    
+    // 輸出 tokens（含思考）
+    this.outputTokens += usage.candidatesTokenCount || 0;
+    this.thoughtTokens += usage.thoughtsTokenCount || 0;
+    
+    // 輸入 tokens（按 modality 分類）
+    if (usage.promptTokensDetails) {
+      for (const detail of usage.promptTokensDetails) {
+        const modality = (detail.modality || 'TEXT').toLowerCase();
+        const count = detail.tokenCount || 0;
+        if (modality === 'audio') {
+          this.inputTokens.audio += count;
+        } else if (modality === 'image') {
+          this.inputTokens.image += count;
+        } else if (modality === 'video') {
+          this.inputTokens.video += count;
+        } else {
+          this.inputTokens.text += count;
+        }
+      }
+    } else {
+      // fallback：沒有 details 時全部算 text
+      this.inputTokens.text += usage.promptTokenCount || 0;
+    }
+  }
+  
+  // 計算費用（美元）
+  calculateCost() {
+    const inputCost = 
+      (this.inputTokens.text + this.inputTokens.image + this.inputTokens.video) / 1_000_000 * PRICING.input.text +
+      this.inputTokens.audio / 1_000_000 * PRICING.input.audio;
+    
+    // 輸出費用包含思考 tokens
+    const outputCost = (this.outputTokens + this.thoughtTokens) / 1_000_000 * PRICING.output;
+    
+    return {
+      inputCost,
+      outputCost,
+      totalCost: inputCost + outputCost
+    };
+  }
+  
+  // 取得摘要
+  getSummary() {
+    const cost = this.calculateCost();
+    const totalInput = this.inputTokens.text + this.inputTokens.audio + this.inputTokens.image + this.inputTokens.video;
+    const totalOutput = this.outputTokens + this.thoughtTokens;
+    
+    return {
+      apiCalls: this.apiCalls,
+      tokens: {
+        input: {
+          text: this.inputTokens.text,
+          audio: this.inputTokens.audio,
+          image: this.inputTokens.image,
+          video: this.inputTokens.video,
+          total: totalInput
+        },
+        output: this.outputTokens,
+        thought: this.thoughtTokens,
+        totalOutput: totalOutput,
+        total: totalInput + totalOutput
+      },
+      cost: {
+        input: cost.inputCost,
+        output: cost.outputCost,
+        total: cost.totalCost,
+        formatted: `$${cost.totalCost.toFixed(6)}`
+      }
+    };
+  }
+  
+  // 格式化輸出（用於顯示）
+  formatSummary() {
+    const summary = this.getSummary();
+    let lines = [];
+    
+    lines.push(`📊 **API 使用量統計**`);
+    lines.push(`- API 呼叫次數：${summary.apiCalls} 次`);
+    lines.push(`- 輸入 Tokens：${summary.tokens.input.total.toLocaleString()}`);
+    if (summary.tokens.input.audio > 0) {
+      lines.push(`  - 音訊：${summary.tokens.input.audio.toLocaleString()}`);
+    }
+    if (summary.tokens.input.text > 0) {
+      lines.push(`  - 文字：${summary.tokens.input.text.toLocaleString()}`);
+    }
+    lines.push(`- 輸出 Tokens：${summary.tokens.totalOutput.toLocaleString()}`);
+    if (summary.tokens.thought > 0) {
+      lines.push(`  - 思考：${summary.tokens.thought.toLocaleString()}`);
+      lines.push(`  - 回應：${summary.tokens.output.toLocaleString()}`);
+    }
+    lines.push(`- 總 Tokens：${summary.tokens.total.toLocaleString()}`);
+    lines.push('');
+    lines.push(`💰 **預估費用**`);
+    lines.push(`- 輸入費用：$${summary.cost.input.toFixed(6)}`);
+    lines.push(`- 輸出費用：$${summary.cost.output.toFixed(6)}`);
+    lines.push(`- **總費用：${summary.cost.formatted}**`);
+    
+    return lines.join('\n');
+  }
+}
+
 /**
  * 計算動態 timeout
  */
@@ -62,6 +194,48 @@ function withTimeout(promise, ms, message = 'Operation timed out') {
   ]);
 }
 
+/**
+ * 檢查路徑是否包含非 ASCII 字元
+ */
+function hasNonAscii(str) {
+  return /[^\x00-\x7F]/.test(str);
+}
+
+/**
+ * 安全上傳檔案（處理非 ASCII 路徑）
+ * @google/genai SDK 無法處理包含中文字元的檔案路徑
+ */
+async function safeUploadFile(ai, filePath, mimeType, timeout) {
+  let tempPath = null;
+  let actualPath = filePath;
+  
+  // 如果路徑包含非 ASCII 字元，複製到臨時路徑
+  if (hasNonAscii(filePath)) {
+    const tempDir = path.join(tmpdir(), `upload_${Date.now()}`);
+    await fs.mkdir(tempDir, { recursive: true });
+    const ext = path.extname(filePath);
+    tempPath = path.join(tempDir, `audio_${createHash('md5').update(filePath).digest('hex').slice(0, 8)}${ext}`);
+    await fs.copyFile(filePath, tempPath);
+    actualPath = tempPath;
+    console.error(`   📋 已複製到臨時路徑（避免中文路徑問題）`);
+  }
+  
+  try {
+    const uploadedFile = await withTimeout(
+      ai.files.upload({ file: actualPath, config: { mimeType } }),
+      timeout,
+      `上傳超時 (${timeout/1000}s)`
+    );
+    return { uploadedFile, tempPath };
+  } catch (error) {
+    // 清理臨時檔案
+    if (tempPath) {
+      await fs.rm(path.dirname(tempPath), { recursive: true, force: true }).catch(() => {});
+    }
+    throw error;
+  }
+}
+
 /**
  * 取得音檔長度（秒）
  */
@@ -467,7 +641,7 @@ function escapeRegex(str) {
 /**
  * 快速模式：簡單轉錄單一片段（不做語者識別）
  */
-async function transcribeSingleSegmentFast(ai, segment, model, maxRetries = MAX_RETRIES) {
+async function transcribeSingleSegmentFast(ai, segment, model, tracker = null, maxRetries = MAX_RETRIES) {
   const { index, path: segmentPath, offset, duration = SEGMENT_DURATION } = segment;
   let uploadedFile = null;
   const segmentTimeout = calculateTimeout(duration);
@@ -486,13 +660,12 @@ async function transcribeSingleSegmentFast(ai, segment, model, maxRetries = MAX_
 
 開始：`;
 
+  let uploadTempPath = null;
   for (let attempt = 0; attempt < maxRetries; attempt++) {
     try {
-      uploadedFile = await withTimeout(
-        ai.files.upload({ file: segmentPath, config: { mimeType: 'audio/mpeg' } }),
-        segmentTimeout,
-        `上傳超時 (${segmentTimeout/1000}s)`
-      );
+      const uploadResult = await safeUploadFile(ai, segmentPath, 'audio/mpeg', segmentTimeout);
+      uploadedFile = uploadResult.uploadedFile;
+      uploadTempPath = uploadResult.tempPath;
 
       const response = await withTimeout(
         ai.models.generateContent({
@@ -507,7 +680,13 @@ async function transcribeSingleSegmentFast(ai, segment, model, maxRetries = MAX_
         `轉錄超時 (${segmentTimeout/1000}s)`
       );
 
+      // 追蹤 token 用量
+      if (tracker) tracker.addFromResponse(response);
+
       ai.files.delete({ name: uploadedFile.name }).catch(() => {});
+      if (uploadTempPath) {
+        await fs.rm(path.dirname(uploadTempPath), { recursive: true, force: true }).catch(() => {});
+      }
 
       let transcript = response.text;
 
@@ -524,6 +703,10 @@ async function transcribeSingleSegmentFast(ai, segment, model, maxRetries = MAX_
         ai.files.delete({ name: uploadedFile.name }).catch(() => {});
         uploadedFile = null;
       }
+      if (uploadTempPath) {
+        await fs.rm(path.dirname(uploadTempPath), { recursive: true, force: true }).catch(() => {});
+        uploadTempPath = null;
+      }
 
       if (attempt < maxRetries - 1) {
         const delay = RETRY_BASE_DELAY * Math.pow(2, attempt);
@@ -541,7 +724,7 @@ async function transcribeSingleSegmentFast(ai, segment, model, maxRetries = MAX_
  * 語者識別模式：單一 API 呼叫同時轉錄 + 語者識別
  * 回傳格式包含逐字稿和語者資訊，減少 API 呼叫次數
  */
-async function transcribeSingleSegmentSpeaker(ai, segment, model, speakerProfiles = {}, previousEnding = '', maxRetries = MAX_RETRIES) {
+async function transcribeSingleSegmentSpeaker(ai, segment, model, speakerProfiles = {}, previousEnding = '', tracker = null, maxRetries = MAX_RETRIES) {
   const { index, path: segmentPath, offset, duration = SEGMENT_DURATION } = segment;
   let uploadedFile = null;
   const segmentTimeout = calculateTimeout(duration);
@@ -596,13 +779,12 @@ ${previousEnding.slice(-400)}
 
 開始轉錄：`;
 
+  let uploadTempPath = null;
   for (let attempt = 0; attempt < maxRetries; attempt++) {
     try {
-      uploadedFile = await withTimeout(
-        ai.files.upload({ file: segmentPath, config: { mimeType: 'audio/mpeg' } }),
-        segmentTimeout,
-        `上傳超時 (${segmentTimeout/1000}s)`
-      );
+      const uploadResult = await safeUploadFile(ai, segmentPath, 'audio/mpeg', segmentTimeout);
+      uploadedFile = uploadResult.uploadedFile;
+      uploadTempPath = uploadResult.tempPath;
 
       const response = await withTimeout(
         ai.models.generateContent({
@@ -617,7 +799,13 @@ ${previousEnding.slice(-400)}
         `轉錄超時 (${segmentTimeout/1000}s)`
       );
 
+      // 追蹤 token 用量
+      if (tracker) tracker.addFromResponse(response);
+
       ai.files.delete({ name: uploadedFile.name }).catch(() => {});
+      if (uploadTempPath) {
+        await fs.rm(path.dirname(uploadTempPath), { recursive: true, force: true }).catch(() => {});
+      }
 
       const fullText = response.text;
       
@@ -682,7 +870,7 @@ ${previousEnding.slice(-400)}
       return { 
         index, 
         transcript, 
-        emotion,  // 新增
+        emotion,
         success: true, 
         speakerProfiles: newSpeakerProfiles 
       };
@@ -692,6 +880,10 @@ ${previousEnding.slice(-400)}
         ai.files.delete({ name: uploadedFile.name }).catch(() => {});
         uploadedFile = null;
       }
+      if (uploadTempPath) {
+        await fs.rm(path.dirname(uploadTempPath), { recursive: true, force: true }).catch(() => {});
+        uploadTempPath = null;
+      }
 
       if (attempt < maxRetries - 1) {
         const delay = RETRY_BASE_DELAY * Math.pow(2, attempt);
@@ -1103,6 +1295,7 @@ ${transcript}
 async function transcribeAudioFast(audioPath, outputPath, model, ai, durationMin) {
   const segments = await splitAudioFast(audioPath);
   let tempDir = null;
+  const tracker = new UsageTracker();
 
   if (segments.length > 1 && !segments[0].isOriginal) {
     tempDir = path.dirname(segments[0].path);
@@ -1112,10 +1305,7 @@ async function transcribeAudioFast(audioPath, outputPath, model, ai, durationMin
 
   if (segments.length === 1) {
     console.error('⬆️  上傳中...');
-    const uploadedFile = await ai.files.upload({
-      file: audioPath,
-      config: { mimeType: 'audio/mpeg' }
-    });
+    const { uploadedFile, tempPath: uploadTempPath } = await safeUploadFile(ai, audioPath, 'audio/mpeg', 120000);
     console.error('✅ 上傳完成');
 
     const response = await ai.models.generateContent({
@@ -1129,8 +1319,12 @@ async function transcribeAudioFast(audioPath, outputPath, model, ai, durationMin
       config: { maxOutputTokens: 65536, temperature: 0.1 }
     });
 
+    tracker.addFromResponse(response);
     transcript = response.text;
     try { await ai.files.delete({ name: uploadedFile.name }); } catch {}
+    if (uploadTempPath) {
+      await fs.rm(path.dirname(uploadTempPath), { recursive: true, force: true }).catch(() => {});
+    }
 
   } else {
     // 平行處理
@@ -1139,7 +1333,7 @@ async function transcribeAudioFast(audioPath, outputPath, model, ai, durationMin
 
     for (let i = 0; i < segments.length; i += MAX_WORKERS) {
       const batch = segments.slice(i, i + MAX_WORKERS);
-      const batchPromises = batch.map(seg => transcribeSingleSegmentFast(ai, seg, model));
+      const batchPromises = batch.map(seg => transcribeSingleSegmentFast(ai, seg, model, tracker));
       const batchResults = await Promise.all(batchPromises);
       
       for (const r of batchResults) {
@@ -1158,6 +1352,10 @@ async function transcribeAudioFast(audioPath, outputPath, model, ai, durationMin
     await fs.rm(tempDir, { recursive: true, force: true });
   }
 
+  // 取得使用量摘要
+  const usage = tracker.getSummary();
+  console.error(`💰 費用：${usage.cost.formatted}（${usage.tokens.total.toLocaleString()} tokens）`);
+
   // 儲存結果
   const outPath = outputPath || audioPath.replace(/\.[^/.]+$/, '.md');
   const formattedTimestamp = new Date().toISOString().slice(0, 19).replace(/:/g, '-');
@@ -1188,7 +1386,8 @@ ${transcript}
     duration: `${Math.round(durationMin)} 分鐘`,
     segments: segments.length,
     summary: '（快速模式不產生摘要）',
-    keywords: ''
+    keywords: '',
+    usage
   };
 }
 
@@ -1198,6 +1397,7 @@ ${transcript}
 async function transcribeAudioSpeaker(audioPath, outputPath, model, ai, durationMin) {
   const segments = await splitAudioSpeaker(audioPath);
   let tempDir = null;
+  const tracker = new UsageTracker();
 
   if (segments.length > 1 && !segments[0].isOriginal) {
     tempDir = path.dirname(segments[0].path);
@@ -1228,7 +1428,7 @@ async function transcribeAudioSpeaker(audioPath, outputPath, model, ai, duration
 
     // 單一 API 呼叫：同時取得逐字稿 + 情緒分析 + 語者資訊
     const result = await transcribeSingleSegmentSpeaker(
-      ai, segment, model, speakerProfiles, previousEnding
+      ai, segment, model, speakerProfiles, previousEnding, tracker
     );
 
     if (!result.success) {
@@ -1278,6 +1478,7 @@ async function transcribeAudioSpeaker(audioPath, outputPath, model, ai, duration
     ]),
     config: { maxOutputTokens: 2048, temperature: 0.3 }
   });
+  tracker.addFromResponse(summaryResponse);
   const summary = summaryResponse.text;
 
   // 生成簡短情緒摘要（一段文字，包含經典句子）
@@ -1304,6 +1505,7 @@ ${emotionResults.map(r => `[${r.time}分鐘] ${r.emotion.slice(0, 200)}`).join('
         contents: createUserContent([emotionSummaryPrompt]),
         config: { maxOutputTokens: 1024, temperature: 0.3 }
       });
+      tracker.addFromResponse(emotionSummaryResponse);
       emotionSummary = emotionSummaryResponse.text;
       console.error('   ✅ 情緒摘要完成');
     } catch (error) {
@@ -1409,13 +1611,18 @@ ${finalTranscript}
   await fs.writeFile(outPath, finalContent, 'utf8');
   console.error(`📄 [語者識別模式] 逐字稿已儲存: ${outPath}`);
 
+  // 取得使用量摘要
+  const usage = tracker.getSummary();
+  console.error(`💰 費用：${usage.cost.formatted}（${usage.tokens.total.toLocaleString()} tokens）`);
+
   return {
     outputPath: outPath,
     duration: metadata.duration,
     segments: segments.length,
     summary,
     keywords: '',
-    speakerProfiles
+    speakerProfiles,
+    usage
   };
 }
 
@@ -1612,10 +1819,7 @@ async function describeImage(imagePath, outputPath = null, detailLevel = 'detail
 
   const prompt = detailPrompts[detailLevel] || detailPrompts.detailed;
 
-  const uploadedFile = await ai.files.upload({
-    file: imagePath,
-    config: { mimeType: 'image/jpeg' }
-  });
+  const { uploadedFile, tempPath: uploadTempPath } = await safeUploadFile(ai, imagePath, 'image/jpeg', 60000);
 
   const response = await ai.models.generateContent({
     model: MODEL_NAME,
@@ -1627,6 +1831,9 @@ async function describeImage(imagePath, outputPath = null, detailLevel = 'detail
   });
 
   try { await ai.files.delete({ name: uploadedFile.name }); } catch { /* ignore */ }
+  if (uploadTempPath) {
+    await fs.rm(path.dirname(uploadTempPath), { recursive: true, force: true }).catch(() => {});
+  }
 
   const description = response.text;
   const outPath = outputPath || imagePath.replace(/\.[^/.]+$/, '.md');
@@ -1687,43 +1894,36 @@ server.tool(
 
       // 根據模式顯示不同結果（從結果中取得實際使用的模式）
       const actualMode = result.mode || (mode || process.env.DEFAULT_MODE || 'fast');
-      if (actualMode === 'fast') {
-        return {
-          content: [{
-            type: 'text',
-            text: `✅ [快速模式] 音訊轉錄完成！
+      
+      // 格式化費用資訊（簡潔版）
+      const usageInfo = result.usage ? `
 
-📁 **輸出檔案**: ${result.outputPath}
-⏱️ **總時長**: ${result.duration}
-📊 **處理區塊**: ${result.segments} 個`
-          }]
-        };
-      } else if (selectedMode === 'hybrid') {
-        const speakerInfo = result.speakerProfiles ? 
-          Object.entries(result.speakerProfiles)
-            .map(([name, info]) => `- **${name}**（${[info.gender, info.role].filter(Boolean).join('，') || '?'}）`)
-            .join('\n') : '';
+💰 **費用統計**
+- 輸入 Tokens：${result.usage.tokens.input.total.toLocaleString()}
+- 輸出 Tokens：${result.usage.tokens.totalOutput.toLocaleString()}
+- **總費用：${result.usage.cost.formatted}**` : '';
 
+      if (actualMode === 'fast') {
         return {
           content: [{
             type: 'text',
-            text: `✅ [Hybrid 模式] 音訊轉錄完成！
+            text: `✅ [快速模式] 音訊轉錄完成！
 
 📁 **輸出檔案**: ${result.outputPath}
 ⏱️ **總時長**: ${result.duration}
 📊 **處理區塊**: ${result.segments} 個
-
-## 語者資訊
-${speakerInfo || '（未識別）'}
-
-## 摘要
-${result.summary}`
+${usageInfo}`
           }]
         };
       } else {
+        // speaker 模式
         const speakerInfo = result.speakerProfiles ? 
           Object.entries(result.speakerProfiles)
-            .map(([id, info]) => `- ${info.name || id}（${info.role || '?'}）`)
+            .map(([id, info]) => {
+              const name = info.name || id;
+              const details = [info.gender, info.role, info.traits].filter(Boolean).join('，');
+              return `- **${name}**：${details || '待識別'}`;
+            })
             .join('\n') : '';
 
         return {
@@ -1739,7 +1939,8 @@ ${result.summary}`
 ${speakerInfo || '（未識別）'}
 
 ## 摘要
-${result.summary}`
+${result.summary || '（無）'}
+${usageInfo}`
           }]
         };
       }
@@ -1849,11 +2050,8 @@ server.tool(
         ffmpeg.on('error', reject);
       });
 
-      // 上傳並分析
-      const uploadedFile = await ai.files.upload({
-        file: segmentPath,
-        config: { mimeType: 'audio/mpeg' }
-      });
+      // 上傳並分析（使用 safeUploadFile 處理中文路徑）
+      const { uploadedFile, tempPath: uploadTempPath } = await safeUploadFile(ai, segmentPath, 'audio/mpeg', 60000);
 
       const defaultQuestion = `分析這段音頻中說話的人：
 
@@ -1876,6 +2074,9 @@ server.tool(
 
       // 清理
       ai.files.delete({ name: uploadedFile.name }).catch(() => {});
+      if (uploadTempPath) {
+        await fs.rm(path.dirname(uploadTempPath), { recursive: true, force: true }).catch(() => {});
+      }
       await fs.rm(tempDir, { recursive: true, force: true });
 
       const timeRange = `${timestamp || '00:00'} - ${formatTime(startSeconds + duration)}`;