feat: 将视频转写重构为纯文字异步语音识别

lib/stt/audio.ts

@@ -0,0 +1,85 @@
+import { execFile } from "node:child_process";
+import { createWriteStream } from "node:fs";
+import { mkdtemp, rm } from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import { pipeline } from "node:stream/promises";
+import { promisify } from "node:util";
+import { extractAudio } from "../../app/api/media";
+import { getFileStream, getFileUrl, uploadTranscriptionAudio } from "../minio";
+import { prisma } from "../prisma";
+import { AsrError, MAX_DURATION_MS, validateMedia } from "./dashscope";
+
+const execFileAsync = promisify(execFile);
+
+// Whole-file transcription: download to disk, extract once, upload only audio.
+export async function prepareAudio(videoId: string): Promise<string | null> {
+  const video = await prisma.video.findUniqueOrThrow({
+    where: { aweme_id: videoId },
+  });
+  if (video.duration_ms > MAX_DURATION_MS)
+    throw new AsrError("视频超过 ASR 的 12 小时时长限制");
+  const dir = await mkdtemp(path.join(os.tmpdir(), "douyin-asr-"));
+  try {
+    const input = path.join(dir, "video.mp4");
+    await pipeline(
+      await getFileStream(video.video_url),
+      createWriteStream(input),
+      { signal: AbortSignal.timeout(60 * 60 * 1000) },
+    );
+    const { stdout } = await execFileAsync(
+      process.env.FFPROBE_PATH || "ffprobe",
+      [
+        "-v",
+        "error",
+        "-select_streams",
+        "a:0",
+        "-show_entries",
+        "stream=index:format=duration",
+        "-of",
+        "json",
+        input,
+      ],
+      { timeout: 60_000 },
+    );
+    const probe = JSON.parse(stdout) as {
+      streams?: { index: number }[];
+      format?: { duration?: string };
+    };
+    if (!Array.isArray(probe.streams)) throw new AsrError("无法读取视频音轨");
+    if (!probe.streams.length) return null;
+    const duration = Number(probe.format?.duration) * 1000;
+    if (!Number.isFinite(duration) || duration <= 0)
+      throw new AsrError("无法读取音频时长");
+    if (duration > MAX_DURATION_MS)
+      throw new AsrError("音频超过 ASR 的 12 小时时长限制");
+    const audio = await extractAudio(input, { format: "mp3", bitrateKbps: 64 });
+    if (!audio) throw new AsrError("未找到 ffmpeg，请安装或配置 FFMPEG_PATH");
+    validateMedia(duration, audio.buffer.length);
+    const objectPath = `transcription-audio/${encodeURIComponent(videoId)}/${Date.now()}.mp3`;
+    await uploadTranscriptionAudio(objectPath, audio.buffer);
+    const url = getFileUrl(objectPath);
+    let parsed: URL;
+    try {
+      parsed = new URL(url);
+    } catch {
+      throw new AsrError("请配置公网可访问的 MINIO_PUBLIC_DOMAIN");
+    }
+    if (!["http:", "https:"].includes(parsed.protocol))
+      throw new AsrError("音频 URL 协议无效");
+    const response = await fetch(url, {
+      method: "HEAD",
+      signal: AbortSignal.timeout(30_000),
+    });
+    if (!response.ok)
+      throw new AsrError(`音频公共地址不可访问 (HTTP ${response.status})`);
+    return url;
+  } catch (error) {
+    if (error instanceof AsrError) throw error;
+    throw new AsrError(
+      "音频准备失败，请检查 MinIO、ffmpeg、ffprobe 和文件可访问性后重试",
+    );
+  } finally {
+    await rm(dir, { recursive: true, force: true });
+  }
+}