11

scripts/nodejs/pipeline_native.js

@@ -294,9 +294,109 @@ function resolveTtsModel(voiceId) {
   return "qwen3-tts-flash";
 }
 
+function splitTextForTts(text, maxLen = 300) {
+  const t = String(text || "").trim();
+  if (t.length <= maxLen) return [t];
+  const segments = [];
+  let remaining = t;
+  while (remaining.length > 0) {
+    if (remaining.length <= maxLen) {
+      segments.push(remaining.trim());
+      break;
+    }
+    let splitIdx = -1;
+    for (const sep of ["。", "！", "？", ".", "!", "?"]) {
+      const idx = remaining.lastIndexOf(sep, maxLen);
+      if (idx > maxLen * 0.3) {
+        splitIdx = idx + 1;
+        break;
+      }
+    }
+    if (splitIdx === -1) {
+      for (const sep of ["，", ",", "、", "：", ":"]) {
+        const idx = remaining.lastIndexOf(sep, maxLen);
+        if (idx > maxLen * 0.3) {
+          splitIdx = idx + 1;
+          break;
+        }
+      }
+    }
+    if (splitIdx === -1) splitIdx = maxLen;
+    segments.push(remaining.slice(0, splitIdx).trim());
+    remaining = remaining.slice(splitIdx).trim();
+  }
+  return segments.filter((s) => s.length > 0);
+}
+
+function concatWavFiles(inputPaths, outputPath) {
+  if (inputPaths.length === 0) throw new Error("无输入文件");
+  if (inputPaths.length === 1) {
+    fs.copyFileSync(inputPaths[0], outputPath);
+    return;
+  }
+  const firstFile = fs.readFileSync(inputPaths[0]);
+  const headerSize = 44;
+  const header = Buffer.from(firstFile.buffer, 0, headerSize);
+  const pcmBuffers = [];
+  let totalPcmSize = 0;
+  for (const p of inputPaths) {
+    const buf = fs.readFileSync(p);
+    const pcm = buf.slice(headerSize);
+    pcmBuffers.push(pcm);
+    totalPcmSize += pcm.length;
+  }
+  const newHeader = Buffer.from(header);
+  newHeader.writeUInt32LE(totalPcmSize + headerSize - 8, 4);
+  newHeader.writeUInt32LE(totalPcmSize, 40);
+  const outputBuf = Buffer.concat([newHeader, ...pcmBuffers]);
+  fs.writeFileSync(outputPath, outputBuf);
+}
+
 /**
- * 通义千问 TTS 合成（对齐 Electron synthesizeOneSegment）
- * @param {{ apiKey: string, text: string, voice: string, model?: string, instruction?: string, outputPath: string }} opts
+ * 通义千问 TTS 单段合成（对齐 Electron synthesizeOneSegment）
+ */
+async function qwenTtsSynthesizeOneSegment(text, config, outputPath) {
+  const apiKey = config.apiKey;
+  const voice = String(config.voice || "").trim();
+  const model = config.model || resolveTtsModel(voice);
+  const instruction = String(config.instruction || "").trim();
+  const languageType = config.languageType;
+
+  const input = {
+    text: String(text || "").trim(),
+    voice,
+    ...(languageType ? { language_type: languageType } : {}),
+    ...(instruction
+      ? { instructions: instruction, optimize_instructions: true }
+      : {}),
+  };
+
+  const data = await fetchJson(QWEN_TTS_URL, {
+    method: "POST",
+    headers: {
+      Authorization: `Bearer ${apiKey}`,
+      "Content-Type": "application/json",
+    },
+    body: JSON.stringify({ model, input }),
+  });
+
+  const audioUrl = data?.output?.audio?.url;
+  if (!audioUrl) {
+    return {
+      success: false,
+      error: data?.message || data?.code || "未获取到音频 URL",
+    };
+  }
+
+  const buf = await fetchArrayBuffer(audioUrl);
+  fs.mkdirSync(path.dirname(outputPath), { recursive: true });
+  fs.writeFileSync(outputPath, buf);
+  return { success: true, audioPath: outputPath };
+}
+
+/**
+ * 通义千问 TTS 合成（支持长文本分段，对齐 Electron synthesizeSpeech）
+ * @param {{ apiKey: string, text: string, voice: string, model?: string, instruction?: string, languageType?: string, outputPath: string }} opts
  */
 async function qwenTtsSynthesize(opts) {
   const apiKey = opts.apiKey;
@@ -305,43 +405,72 @@ async function qwenTtsSynthesize(opts) {
   const outputPath = opts.outputPath;
   const model = opts.model || resolveTtsModel(voice);
   const instruction = String(opts.instruction || "").trim();
+  const languageType = opts.languageType;
 
   if (!apiKey) return { success: false, error: "缺少 DashScope apiKey" };
   if (!text) return { success: false, error: "缺少合成文本" };
   if (!voice) return { success: false, error: "缺少 voice" };
   if (!outputPath) return { success: false, error: "缺少 outputPath" };
 
-  try {
-    const data = await fetchJson(QWEN_TTS_URL, {
-      method: "POST",
-      headers: {
-        Authorization: `Bearer ${apiKey}`,
-        "Content-Type": "application/json",
-      },
-      body: JSON.stringify({
-        model,
-        input: {
-          text,
-          voice,
-          ...(instruction
-            ? { instructions: instruction, optimize_instructions: true }
-            : {}),
-        },
-      }),
-    });
+  const instrLen = instruction.length;
+  const isVcModel = model.includes("-vc") || model.includes("_vc");
+  const baseMaxLen = isVcModel ? 200 : 300;
+  const effectiveMaxLen = Math.max(100, baseMaxLen - instrLen);
+  const segments = splitTextForTts(text, effectiveMaxLen);
 
-    const audioUrl = data?.output?.audio?.url;
-    if (!audioUrl) {
-      return {
-        success: false,
-        error: data?.message || data?.code || "未获取到音频 URL",
-      };
+  const config = { apiKey, voice, model, instruction, languageType };
+
+  try {
+    if (segments.length === 1) {
+      const r = await qwenTtsSynthesizeOneSegment(segments[0], config, outputPath);
+      if (!r.success) return r;
+      log("info", "qwen tts saved", { outputPath, voice, model });
+      return { success: true, audioPath: outputPath };
     }
 
-    const buf = await fetchArrayBuffer(audioUrl);
-    fs.mkdirSync(path.dirname(outputPath), { recursive: true });
-    fs.writeFileSync(outputPath, buf);
-    log("info", "qwen tts saved", { outputPath, voice, model });
+    const parsed = path.parse(outputPath);
+    const ext = parsed.ext || ".wav";
+    const segmentPaths = [];
+    for (let i = 0; i < segments.length; i++) {
+      const segPath = path.join(
+        parsed.dir,
+        `${parsed.name}_seg${i}${ext}`,
+      );
+      log("info", "qwen tts segment", { i: i + 1, total: segments.length, len: segments[i].length });
+      const segResult = await qwenTtsSynthesizeOneSegment(segments[i], config, segPath);
+      if (!segResult.success) return segResult;
+      segmentPaths.push(segPath);
+    }
+
+    const outExt = ext.toLowerCase();
+    if (outExt === ".wav") {
+      concatWavFiles(segmentPaths, outputPath);
+    } else {
+      const wavOut = path.join(parsed.dir, `${parsed.name}_merged.wav`);
+      concatWavFiles(segmentPaths, wavOut);
+      try {
+        await runCommand(exeName(), [
+          "-y",
+          "-i",
+          wavOut,
+          "-c:a",
+          "libmp3lame",
+          "-b:a",
+          "128k",
+          outputPath,
+        ]);
+        deleteFile(wavOut);
+      } catch (e) {
+        fs.copyFileSync(wavOut, outputPath.replace(/\.mp3$/i, ".wav"));
+        return {
+          success: false,
+          error: `分段合成成功但转 mp3 失败: ${e.message}（已保存 wav）`,
+        };
+      }
+    }
+
+    for (const p of segmentPaths) deleteFile(p);
+    log("info", "qwen tts merged", { outputPath, segments: segments.length });
     return { success: true, audioPath: outputPath };
   } catch (e) {
     return { success: false, error: e.message || String(e) };
@@ -529,5 +658,6 @@ module.exports = {
   localAsr,
   openaiChat,
   resolveTtsModel,
+  splitTextForTts,
   qwenTtsSynthesize,
 };