feat: 将视频转写重构为纯文字异步语音识别

2026-10-03 21:58:06 +08:00
parent 040273586b
commit 6f0e38054b
34 changed files with 2283 additions and 1771 deletions
@@ -0,0 +1,77 @@
+import "dotenv/config";
+import { writeFile } from "node:fs/promises";
+import { setTimeout as sleep } from "node:timers/promises";
+import { prisma } from "../lib/prisma";
+import { prepareAudio } from "../lib/stt/audio";
+import {
+  downloadTranscript,
+  getAsrConfig,
+  queryTranscription,
+  submitTranscription,
+} from "../lib/stt/dashscope";
+
+async function main() {
+  const videoId = process.argv[2];
+  if (!videoId || !/^[A-Za-z0-9_-]+$/.test(videoId))
+    throw new Error("用法：npx tsx scripts/stt-smoke.ts <视频 ID>");
+  getAsrConfig();
+  const video = await prisma.video.findUniqueOrThrow({
+    where: { aweme_id: videoId },
+    select: { duration_ms: true },
+  });
+  console.log(
+    JSON.stringify({
+      videoId,
+      durationSeconds: video.duration_ms / 1000,
+      phase: "extracting_audio",
+    }),
+  );
+  const audioURL = await prepareAudio(videoId);
+  const resultPath = `/tmp/douyin-asr-${videoId}.json`;
+  if (!audioURL) {
+    await writeFile(
+      resultPath,
+      JSON.stringify({ videoId, transcript: [] }, null, 2),
+    );
+    console.log(
+      JSON.stringify({ phase: "complete", transcript: [], resultPath }),
+    );
+    return;
+  }
+  const taskId = await submitTranscription(audioURL);
+  await writeFile(
+    resultPath,
+    JSON.stringify({ videoId, taskId, status: "submitted" }, null, 2),
+  );
+  console.log(JSON.stringify({ phase: "submitted", taskId }));
+  for (let attempt = 0; attempt < 90; attempt++) {
+    await sleep(10_000);
+    const task = await queryTranscription(taskId);
+    if (task.status === "SUCCEEDED") {
+      const transcript = await downloadTranscript(task.resultURL);
+      await writeFile(
+        resultPath,
+        JSON.stringify({ videoId, taskId, transcript }, null, 2),
+      );
+      console.log(
+        JSON.stringify({ phase: "complete", transcript, resultPath }),
+      );
+      return;
+    }
+    if (attempt % 6 === 0) console.log(JSON.stringify({ phase: task.status }));
+  }
+  throw new Error(
+    `15 分钟内未完成，任务 ID 已保存在 ${resultPath}，请查询原任务而非重复提交`,
+  );
+}
+main()
+  .catch((error) => {
+    // Do not print arbitrary SDK/network errors: they may contain signed URLs.
+    console.error(
+      error instanceof Error && !error.name.startsWith("Prisma")
+        ? error.message
+        : "读取视频或转写失败",
+    );
+    process.exitCode = 1;
+  })
+  .finally(() => prisma.$disconnect());
@@ -0,0 +1,34 @@
+import "dotenv/config";
+import { setTimeout as sleep } from "node:timers/promises";
+import { prisma } from "../lib/prisma";
+import { workerTick } from "../lib/stt/worker";
+
+let stopping = false;
+const controller = new AbortController();
+for (const signal of ["SIGTERM", "SIGINT"] as const)
+  process.on(signal, () => {
+    stopping = true;
+    controller.abort();
+  });
+async function main() {
+  console.log("[ASR] 后台执行器已启动，最大并发 2");
+  try {
+    while (!stopping) {
+      try {
+        await workerTick();
+      } catch {
+        console.error("[ASR] 数据库调度失败，稍后重试");
+      }
+      if (!stopping)
+        await sleep(1000, undefined, { signal: controller.signal }).catch(
+          () => {},
+        );
+    }
+  } finally {
+    await prisma.$disconnect();
+  }
+}
+main().catch(() => {
+  console.error("[ASR] 执行器退出");
+  process.exitCode = 1;
+});