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 +1,22 @@
+import { NextResponse } from "next/server";
+import { getAsrConfig } from "@/lib/stt/dashscope";
+import { enqueueMissingTranscriptions } from "@/lib/stt/tasks";
+export const runtime = "nodejs";
+export async function POST() {
+  try {
+    getAsrConfig();
+  } catch (error) {
+    return NextResponse.json(
+      { error: (error as Error).message },
+      { status: 503 },
+    );
+  }
+  try {
+    return NextResponse.json(
+      { queued: await enqueueMissingTranscriptions() },
+      { status: 202 },
+    );
+  } catch {
+    return NextResponse.json({ error: "批量任务入队失败" }, { status: 500 });
+  }
+}
@@ -1,113 +0,0 @@
-import fs from "fs";
-import OpenAI from "openai";
-import prompt from "./prompt.md";
-import { prisma } from "@/lib/prisma";
-import { downloadFile } from "@/lib/minio";
-import { extractAudio } from "../media";
-import { z } from "zod";
-import { zodResponseFormat } from "openai/helpers/zod";
-import { ProxyAgent, setGlobalDispatcher } from "undici";
-
-const proxy = process.env.HTTPS_PROXY || process.env.HTTP_PROXY;
-if (proxy) {
-  setGlobalDispatcher(new ProxyAgent(proxy));
-}
-
-const client = new OpenAI({
-  apiKey: process.env.OPENAI_API_KEY,
-  baseURL: process.env.OPENAI_API_BASE_URL,
-});
-
-// Zod schema aligned with prompt.md
-export const SttSchema = z
-  .object({
-    speech_detected: z.boolean(),
-    language: z.string().min(1).nullable(),
-    audio_type: z.string().nullable(),
-    transcript: z.array(z.string()),
-    non_speech_summary: z.string().nullable(),
-  })
-  .strict();
-
-export type SttResult = z.infer<typeof SttSchema>;
-
-async function transcriptAudio(audio: Buffer | string) {
-  if (typeof audio === "string") {
-    audio = fs.readFileSync(audio);
-  }
-  const base64Audio = Buffer.from(audio).toString("base64");
-
-  const completion = await client.chat.completions.create({
-    model: "gemini-2.5-flash",
-    messages: [
-      {
-        role: "user",
-        content: [
-          { type: "text", text: prompt },
-          {
-            type: "input_audio",
-            input_audio: { data: base64Audio, format: "mp3" },
-          },
-        ],
-      },
-    ],
-    response_format: zodResponseFormat(SttSchema, "stt_result"),
-  });
-
-  const data = completion.choices?.[0]?.message;
-  console.log("转写结果", data.content);
-
-  if (!data || !data.content) {
-    throw new Error("No STT result returned from model");
-  }
-  try {
-    return JSON.parse(data.content) as SttResult;
-  } catch (e) {
-    throw new Error("Failed to parse STT result JSON");
-  }
-}
-
-export async function transcriptAweme(awemeId: string): Promise<SttResult> {
-  const aweme = await prisma.video.findUnique({
-    where: { aweme_id: awemeId },
-  });
-  if (!aweme) {
-    throw new Error("Aweme not found or aweme is not a video post");
-  }
-  const vPath = aweme.video_url;
-  const buffer = await downloadFile(vPath);
-
-  const audioDat = await extractAudio(buffer, {
-    format: "mp3",
-    bitrateKbps: 128,
-  });
-
-  if (!audioDat || !audioDat.buffer) {
-    throw new Error("Failed to extract audio from video");
-  }
-
-  // 调用大模型生成转写结果
-  const result = await transcriptAudio(audioDat.buffer);
-
-  // 将转写结果持久化到数据库
-  await prisma.videoTranscript.upsert({
-    where: { videoId: awemeId },
-    update: {
-      speech_detected: result.speech_detected,
-      language: result.language,
-      audio_type: result.audio_type,
-      transcript: result.transcript,
-      non_speech_summary: result.non_speech_summary,
-    },
-    create: {
-      videoId: awemeId,
-      speech_detected: result.speech_detected,
-      language: result.language,
-      audio_type: result.audio_type,
-      transcript: result.transcript,
-      non_speech_summary: result.non_speech_summary,
-    },
-  });
-
-  return result;
-}
@@ -1,29 +0,0 @@
-你将接收一段音频。请完成：你将接收一段音频。语音活动检测（VAD）与声源分类。
-
-**输出 JSON Schema（示例）**
-
-```json
-{
-  "speech_detected": true,
-  "language": "zh-CN",
-  "audio_type": null,
-  "transcript": ["大家好，我是xxx", "欢迎来到今天的视频", "今天我们来聊聊AI的未来"],
-  "non_speech_summary": null,
-}
-```
-
-**当无发言时返回：** 
-
-```json
-{
-  "speech_detected": false,
-  "language": null,
-  "audio_type": "music | ambience | animal | mechanical | other",
-  "transcript": [],
-  "non_speech_summary": "示例：纯音乐-钢琴独奏，节奏舒缓；或 环境声-雨声伴随雷鸣。",
-}
-```
-
-** 规则补充 **
-
-* 允许`language` 为多标签（如 "zh-CN,en"）或为 `null`（无发言时）。
@@ -1,23 +1,33 @@
+import { type NextRequest, NextResponse } from "next/server";
+import { enqueueTranscription, readTranscription } from "@/lib/stt/tasks";
+import { isActive } from "@/lib/stt/types";
 export const runtime = "nodejs";
-import { NextRequest, NextResponse } from "next/server";
-import { prisma } from "@/lib/prisma";
-import type { FeedItem, FeedResponse } from "@/app/types/feed";
-import { getFileUrl } from "@/lib/minio";
-import { transcriptAweme } from ".";
-
-// Contract
-// Inputs: search params { before?: ISOString, limit?: number }
-// Output: { items: FeedItem[], nextCursor: ISOString | null }
-
+export const dynamic = "force-dynamic";
+const validId = (id: unknown): id is string =>
+  typeof id === "string" && /^[A-Za-z0-9_-]{1,128}$/.test(id);
 export async function GET(req: NextRequest) {
-  const { searchParams } = new URL(req.url);
-  const awemeId = searchParams.get("awemeId");
-  if (!awemeId) {
-    return NextResponse.json(
-      { error: "Missing awemeId parameter" },
-      { status: 400 },
-    );
+  const id = req.nextUrl.searchParams.get("awemeId");
+  if (!validId(id))
+    return NextResponse.json({ error: "需要有效的 awemeId" }, { status: 400 });
+  try {
+    const state = await readTranscription(id);
+    return state
+      ? NextResponse.json(state)
+      : NextResponse.json({ error: "视频不存在" }, { status: 404 });
+  } catch {
+    return NextResponse.json({ error: "读取转写状态失败" }, { status: 500 });
+  }
+}
+export async function POST(req: NextRequest) {
+  const body = await req.json().catch(() => null);
+  if (!validId(body?.awemeId))
+    return NextResponse.json({ error: "需要有效的 awemeId" }, { status: 400 });
+  try {
+    const state = await enqueueTranscription(body.awemeId, true);
+    return state
+      ? NextResponse.json(state, { status: isActive(state.status) ? 202 : 200 })
+      : NextResponse.json({ error: "视频不存在" }, { status: 404 });
+  } catch {
+    return NextResponse.json({ error: "转写任务入队失败" }, { status: 500 });
  }
-  const script = await transcriptAweme(awemeId);
-  return NextResponse.json(script);
 }