使用 qwen 进行语音识别，加入 tag

Co-authored-by: Copilot <copilot@github.com>

app/admin/asr/page.tsx

@@ -3,26 +3,27 @@
 import Link from "next/link";
 import BackButton from "@/app/components/BackButton";
 
-export default function SttAdminPage() {
+export default function AsrAdminPage() {
   return (
     <div className="min-h-screen bg-gray-50">
       <div className="max-w-4xl mx-auto p-8">
         <BackButton />
 
         <div className="mt-8">
-          <h1 className="text-3xl font-bold text-gray-900">STT 管理中心</h1>
+          <h1 className="text-3xl font-bold text-gray-900">ASR 管理中心</h1>
-          <p className="text-gray-600 mt-2">管理和配置视频语音转写功能</p>
+          <p className="text-gray-600 mt-2">管理和配置视频语音识别功能</p>
         </div>
 
         <div className="mt-8 grid gap-6">
           {/* 视频转写管理 */}
           <Link
-            href="/admin/stt/videos"
+            href="/admin/asr/videos"
             className="block bg-white rounded-lg shadow-md hover:shadow-lg transition-shadow p-6 border border-gray-200 hover:border-blue-500"
           >
             <div className="flex items-start gap-4">
               <div className="flex-shrink-0 w-12 h-12 bg-blue-100 rounded-lg flex items-center justify-center">
                 <svg
+                  aria-hidden="true"
                   className="w-6 h-6 text-blue-600"
                   fill="none"
                   stroke="currentColor"
@@ -46,6 +47,7 @@ export default function SttAdminPage() {
                 <div className="mt-4 text-blue-600 font-medium flex items-center gap-1">
                   进入管理
                   <svg
+                    aria-hidden="true"
                     className="w-4 h-4"
                     fill="none"
                     stroke="currentColor"
@@ -68,6 +70,7 @@ export default function SttAdminPage() {
             <div className="flex items-start gap-4">
               <div className="flex-shrink-0 w-12 h-12 bg-gray-100 rounded-lg flex items-center justify-center">
                 <svg
+                  aria-hidden="true"
                   className="w-6 h-6 text-gray-600"
                   fill="none"
                   stroke="currentColor"
@@ -104,6 +107,7 @@ export default function SttAdminPage() {
             <div className="flex items-start gap-4">
               <div className="flex-shrink-0 w-12 h-12 bg-gray-100 rounded-lg flex items-center justify-center">
                 <svg
+                  aria-hidden="true"
                   className="w-6 h-6 text-gray-600"
                   fill="none"
                   stroke="currentColor"

app/admin/asr/videos/page.tsx

@@ -1,7 +1,8 @@
 "use client";
 
-import { useState, useEffect } from "react";
+import Image from "next/image";
 import Link from "next/link";
+import { useCallback, useEffect, useState } from "react";
 import BackButton from "@/app/components/BackButton";
 
 type VideoTranscript = {
@@ -9,8 +10,9 @@ type VideoTranscript = {
   speech_detected: boolean;
   language: string | null;
   audio_type: string | null;
-  transcript: string | null;
+  transcript: string[];
   non_speech_summary: string | null;
+  content_tags: string[];
 };
 
 type Video = {
@@ -29,25 +31,21 @@ type ApiResponse = {
   total: number;
 };
 
-export default function SttVideosPage() {
+export default function AsrVideosPage() {
   const [videos, setVideos] = useState<Video[]>([]);
   const [loading, setLoading] = useState(true);
   const [error, setError] = useState<string | null>(null);
   const [transcribing, setTranscribing] = useState<Set<string>>(new Set());
-  const [filter, setFilter] = useState<"all" | "transcribed" | "pending">(
+  const [filter, setFilter] = useState<
-    "all",
+    "all" | "transcribed" | "pending" | "tagged" | "missingTags"
-  );
+  >("all");
   const [batchTranscribing, setBatchTranscribing] = useState(false);
   const [batchProgress, setBatchProgress] = useState({ current: 0, total: 0 });
 
-  useEffect(() => {
+  const fetchVideos = useCallback(async () => {
-    fetchVideos();
-  }, []);
-
-  const fetchVideos = async () => {
     try {
       setLoading(true);
-      const response = await fetch("/api/admin/stt/videos");
+      const response = await fetch("/api/admin/asr/videos");
       if (!response.ok) {
         throw new Error("Failed to fetch videos");
       }
@@ -58,34 +56,41 @@ export default function SttVideosPage() {
     } finally {
       setLoading(false);
     }
-  };
+  }, []);
+
+  useEffect(() => {
+    fetchVideos();
+  }, [fetchVideos]);
 
   // 简单的 sleep 工具
   const sleep = (ms: number) =>
     new Promise((resolve) => setTimeout(resolve, ms));
 
+  const hasContentTags = (video: Video) =>
+    (video.transcript?.content_tags.length ?? 0) > 0;
+
   type PollOptions = {
     intervalMs?: number;
     maxAttempts?: number;
   };
 
-  // 轮询某个视频的转写状态直至完成或超时
+  // 轮询某个视频的识别和标签状态直至完成或超时
   const pollVideoUntilTranscribed = async (
     awemeId: string,
     { intervalMs = 2000, maxAttempts = 30 }: PollOptions = {},
   ): Promise<boolean> => {
     for (let attempt = 0; attempt < maxAttempts; attempt++) {
       try {
-        const resp = await fetch("/api/admin/stt/videos");
+        const resp = await fetch("/api/admin/asr/videos");
         if (resp.ok) {
           const data: ApiResponse = await resp.json();
           setVideos(data.videos);
           const v = data.videos.find((x) => x.aweme_id === awemeId);
-          if (v && v.transcript) {
+          if ((v?.transcript?.content_tags.length ?? 0) > 0) {
             return true;
           }
         }
-      } catch (e) {
+      } catch {
         // 忽略网络错误，继续轮询
       }
       await sleep(intervalMs);
@@ -96,7 +101,7 @@ export default function SttVideosPage() {
   const handleTranscribe = async (awemeId: string) => {
     setTranscribing((prev) => new Set(prev).add(awemeId));
     try {
-      const response = await fetch(`/api/stt?awemeId=${awemeId}`);
+      const response = await fetch(`/api/asr?awemeId=${awemeId}`);
       if (!response.ok) {
         throw new Error("Failed to transcribe video");
       }
@@ -118,15 +123,15 @@ export default function SttVideosPage() {
   };
 
   const handleBatchTranscribe = async () => {
-    const pendingVideos = videos.filter((v) => v.transcript === null);
+    const pendingVideos = videos.filter((v) => !hasContentTags(v));
     if (pendingVideos.length === 0) {
-      alert("没有待转写的视频");
+      alert("没有缺失标签的视频");
       return;
     }
 
     if (
       !confirm(
-        `确定要转写 ${pendingVideos.length} 个视频吗？这可能需要较长时间。`,
+        `确定要为 ${pendingVideos.length} 个视频生成 ASR 标签吗？这会重新执行识别流程，可能需要较长时间。`,
       )
     ) {
       return;
@@ -142,7 +147,7 @@ export default function SttVideosPage() {
       setTranscribing((prev) => new Set(prev).add(video.aweme_id));
 
       try {
-        const response = await fetch(`/api/stt?awemeId=${video.aweme_id}`);
+        const response = await fetch(`/api/asr?awemeId=${video.aweme_id}`);
         if (!response.ok) {
           console.error(`Failed to transcribe video ${video.aweme_id}`);
         }
@@ -167,12 +172,14 @@ export default function SttVideosPage() {
     await fetchVideos();
     setBatchTranscribing(false);
     setBatchProgress({ current: 0, total: 0 });
-    alert("批量转写完成！");
+    alert("缺失标签生成完成！");
   };
 
   const filteredVideos = videos.filter((v) => {
     if (filter === "transcribed") return v.transcript !== null;
     if (filter === "pending") return v.transcript === null;
+    if (filter === "tagged") return hasContentTags(v);
+    if (filter === "missingTags") return !hasContentTags(v);
     return true;
   });
 
@@ -180,6 +187,8 @@ export default function SttVideosPage() {
     total: videos.length,
     transcribed: videos.filter((v) => v.transcript !== null).length,
     pending: videos.filter((v) => v.transcript === null).length,
+    tagged: videos.filter((v) => hasContentTags(v)).length,
+    missingTags: videos.filter((v) => !hasContentTags(v)).length,
   };
 
   if (loading) {
@@ -202,6 +211,7 @@ export default function SttVideosPage() {
           <div className="bg-red-50 border border-red-200 rounded-lg p-6 text-center">
             <p className="text-red-600">错误: {error}</p>
             <button
+              type="button"
               onClick={fetchVideos}
               className="mt-4 px-4 py-2 bg-red-600 text-white rounded-lg hover:bg-red-700"
             >
@@ -220,13 +230,13 @@ export default function SttVideosPage() {
         <div className="mb-8">
           <BackButton />
           <h1 className="text-3xl font-bold text-gray-900 mt-4">
-            视频转写管理
+            视频语音识别管理
           </h1>
-          <p className="text-gray-600 mt-2">管理视频的语音转写状态</p>
+          <p className="text-gray-600 mt-2">管理视频的语音识别和自动标签状态</p>
         </div>
 
         {/* Stats */}
-        <div className="grid grid-cols-1 md:grid-cols-3 gap-6 mb-8">
+        <div className="grid grid-cols-1 md:grid-cols-4 gap-6 mb-8">
           <div className="bg-white rounded-lg shadow p-6">
             <div className="text-sm text-gray-600">总视频数</div>
             <div className="text-3xl font-bold text-gray-900 mt-2">
@@ -245,6 +255,15 @@ export default function SttVideosPage() {
               {stats.pending}
             </div>
           </div>
+          <div className="bg-white rounded-lg shadow p-6">
+            <div className="text-sm text-gray-600">已生成标签</div>
+            <div className="text-3xl font-bold text-blue-600 mt-2">
+              {stats.tagged}
+            </div>
+            <div className="mt-1 text-xs text-gray-500">
+              缺失 {stats.missingTags}
+            </div>
+          </div>
         </div>
 
         {/* Filter */}
@@ -252,6 +271,7 @@ export default function SttVideosPage() {
           <div className="flex items-center justify-between">
             <div className="flex gap-4">
               <button
+                type="button"
                 onClick={() => setFilter("all")}
                 className={`px-4 py-2 rounded-lg font-medium transition-colors ${
                   filter === "all"
@@ -262,6 +282,7 @@ export default function SttVideosPage() {
                 全部 ({stats.total})
               </button>
               <button
+                type="button"
                 onClick={() => setFilter("transcribed")}
                 className={`px-4 py-2 rounded-lg font-medium transition-colors ${
                   filter === "transcribed"
@@ -272,6 +293,7 @@ export default function SttVideosPage() {
                 已转写 ({stats.transcribed})
               </button>
               <button
+                type="button"
                 onClick={() => setFilter("pending")}
                 className={`px-4 py-2 rounded-lg font-medium transition-colors ${
                   filter === "pending"
@@ -281,25 +303,48 @@ export default function SttVideosPage() {
               >
                 待转写 ({stats.pending})
               </button>
+              <button
+                type="button"
+                onClick={() => setFilter("tagged")}
+                className={`px-4 py-2 rounded-lg font-medium transition-colors ${
+                  filter === "tagged"
+                    ? "bg-blue-600 text-white"
+                    : "bg-gray-100 text-gray-700 hover:bg-gray-200"
+                }`}
+              >
+                已生成标签 ({stats.tagged})
+              </button>
+              <button
+                type="button"
+                onClick={() => setFilter("missingTags")}
+                className={`px-4 py-2 rounded-lg font-medium transition-colors ${
+                  filter === "missingTags"
+                    ? "bg-rose-600 text-white"
+                    : "bg-gray-100 text-gray-700 hover:bg-gray-200"
+                }`}
+              >
+                缺标签 ({stats.missingTags})
+              </button>
             </div>
             <div className="flex items-center gap-4">
               {batchTranscribing && (
                 <div className="text-sm text-gray-600">
-                  正在转写: {batchProgress.current} / {batchProgress.total}
+                  正在生成: {batchProgress.current} / {batchProgress.total}
                 </div>
               )}
               <button
+                type="button"
                 onClick={handleBatchTranscribe}
-                disabled={batchTranscribing || stats.pending === 0}
+                disabled={batchTranscribing || stats.missingTags === 0}
                 className="px-6 py-2 bg-purple-600 text-white rounded-lg hover:bg-purple-700 disabled:bg-gray-400 disabled:cursor-not-allowed transition-colors font-medium"
               >
                 {batchTranscribing ? (
                   <span className="flex items-center gap-2">
                     <span className="inline-block animate-spin rounded-full h-4 w-4 border-b-2 border-white"></span>
-                    批量转写中...
+                    生成中...
                   </span>
                 ) : (
-                  "全部转写"
+                  "一键生成缺失标签"
                 )}
               </button>
             </div>
@@ -319,7 +364,7 @@ export default function SttVideosPage() {
                     时长
                   </th>
                   <th className="px-6 py-3 text-left text-xs font-medium text-gray-500 uppercase tracking-wider">
-                    转写状态
+                    识别/标签状态
                   </th>
                   <th className="px-6 py-3 text-left text-xs font-medium text-gray-500 uppercase tracking-wider">
                     操作
@@ -331,9 +376,12 @@ export default function SttVideosPage() {
                   <tr key={video.aweme_id} className="hover:bg-gray-50">
                     <td className="px-6 py-4">
                       <div className="flex items-center gap-4">
-                        <img
+                        <Image
                           src={video.cover_url}
                           alt=""
+                          width={80}
+                          height={112}
+                          unoptimized
                           className="w-20 h-28 object-cover rounded"
                         />
                         <div className="flex-1 min-w-0">
@@ -367,7 +415,7 @@ export default function SttVideosPage() {
                           </div>
                           {video.transcript.speech_detected ? (
                             <div className="text-xs text-gray-600 line-clamp-5 max-w-md">
-                              {video.transcript.transcript}
+                              {video.transcript.transcript.join(" ")}
                             </div>
                           ) : (
                             <div className="text-xs text-gray-500">
@@ -381,6 +429,23 @@ export default function SttVideosPage() {
                               )}
                             </div>
                           )}
+                          {video.transcript.content_tags.length > 0 && (
+                            <div className="flex flex-wrap gap-1 pt-1">
+                              {video.transcript.content_tags.map((tag) => (
+                                <span
+                                  key={tag}
+                                  className="inline-flex items-center rounded bg-blue-50 px-2 py-0.5 text-xs font-medium text-blue-700"
+                                >
+                                  {tag}
+                                </span>
+                              ))}
+                            </div>
+                          )}
+                          {video.transcript.content_tags.length === 0 && (
+                            <span className="inline-flex items-center rounded bg-rose-50 px-2 py-0.5 text-xs font-medium text-rose-700">
+                              未生成标签
+                            </span>
+                          )}
                         </div>
                       ) : (
                         <span className="inline-flex items-center px-2.5 py-0.5 rounded-full text-xs font-medium bg-orange-100 text-orange-800">
@@ -391,6 +456,7 @@ export default function SttVideosPage() {
                     <td className="px-6 py-4 whitespace-nowrap text-sm">
                       {video.transcript ? (
                         <button
+                          type="button"
                           onClick={() => handleTranscribe(video.aweme_id)}
                           disabled={transcribing.has(video.aweme_id)}
                           className="px-4 py-2 bg-gray-600 text-white rounded-lg hover:bg-gray-700 disabled:bg-gray-400 disabled:cursor-not-allowed transition-colors"
@@ -398,14 +464,17 @@ export default function SttVideosPage() {
                           {transcribing.has(video.aweme_id) ? (
                             <span className="flex items-center gap-2">
                               <span className="inline-block animate-spin rounded-full h-4 w-4 border-b-2 border-white"></span>
-                              重新转写...
+                              生成中...
                             </span>
+                          ) : video.transcript.content_tags.length === 0 ? (
+                            "生成标签"
                           ) : (
-                            "重新转写"
+                            "重新识别"
                           )}
                         </button>
                       ) : (
                         <button
+                          type="button"
                           onClick={() => handleTranscribe(video.aweme_id)}
                           disabled={transcribing.has(video.aweme_id)}
                           className="px-4 py-2 bg-blue-600 text-white rounded-lg hover:bg-blue-700 disabled:bg-blue-400 disabled:cursor-not-allowed transition-colors"
@@ -413,10 +482,10 @@ export default function SttVideosPage() {
                           {transcribing.has(video.aweme_id) ? (
                             <span className="flex items-center gap-2">
                               <span className="inline-block animate-spin rounded-full h-4 w-4 border-b-2 border-white"></span>
-                              转写中...
+                              识别中...
                             </span>
                           ) : (
-                            "开始转写"
+                            "开始识别"
                           )}
                         </button>
                       )}

app/api/admin/asr/videos/route.ts

@@ -1,6 +1,6 @@
 import { NextResponse } from "next/server";
-import { prisma } from "@/lib/prisma";
 import { getFileUrl } from "@/lib/minio";
+import { prisma } from "@/lib/prisma";
 
 export async function GET() {
   try {
@@ -21,6 +21,7 @@ export async function GET() {
             audio_type: true,
             transcript: true,
             non_speech_summary: true,
+            content_tags: true,
           },
         },
       },
@@ -42,6 +43,7 @@ export async function GET() {
             audio_type: v.transcript.audio_type,
             transcript: v.transcript.transcript,
             non_speech_summary: v.transcript.non_speech_summary,
+            content_tags: v.transcript.content_tags,
           }
         : null,
     }));

app/api/asr/index.ts

@@ -0,0 +1,227 @@
+import OpenAI from "openai";
+import { z } from "zod";
+import { ASR_CONTENT_TAGS, normalizeAsrContentTags } from "@/lib/asr-tags";
+import { downloadFile } from "@/lib/minio";
+import { prisma } from "@/lib/prisma";
+import {
+  type ExtractedVideoFrame,
+  extractAudio,
+  extractVideoFrames,
+} from "../media";
+import prompt from "./prompt.md";
+
+export type AsrLogger = (message: string) => void;
+
+type TranscriptAwemeOptions = {
+  log?: AsrLogger;
+};
+
+let qwenClient: OpenAI | null = null;
+
+function getQwenClient() {
+  if (qwenClient) return qwenClient;
+
+  const apiKey = process.env.QWEN_API_KEY || process.env.DASHSCOPE_API_KEY;
+  if (!apiKey) {
+    throw new Error("Missing QWEN_API_KEY environment variable");
+  }
+
+  qwenClient = new OpenAI({
+    apiKey,
+    baseURL:
+      process.env.QWEN_API_BASE_URL ||
+      "https://dashscope.aliyuncs.com/compatible-mode/v1",
+  });
+
+  return qwenClient;
+}
+
+// Zod schema aligned with prompt.md
+export const AsrSchema = 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(),
+    content_tags: z.array(z.enum(ASR_CONTENT_TAGS)).min(1).max(3),
+  })
+  .strict();
+
+const RawAsrSchema = 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(),
+  content_tags: z.unknown().optional(),
+});
+
+export type AsrResult = z.infer<typeof AsrSchema>;
+
+function extractJsonObject(text: string) {
+  const trimmed = text.trim();
+  const fenced = trimmed.match(/```(?:json)?\s*([\s\S]*?)\s*```/i);
+  if (fenced?.[1]) return fenced[1].trim();
+
+  const start = trimmed.indexOf("{");
+  const end = trimmed.lastIndexOf("}");
+  if (start >= 0 && end > start) {
+    return trimmed.slice(start, end + 1);
+  }
+
+  return trimmed;
+}
+
+function parseAsrResult(content: string): AsrResult {
+  const parsed = RawAsrSchema.parse(JSON.parse(extractJsonObject(content)));
+  return AsrSchema.parse({
+    ...parsed,
+    content_tags: normalizeAsrContentTags(parsed.content_tags),
+  });
+}
+
+async function transcriptAudio(
+  audio: Buffer,
+  frames: ExtractedVideoFrame[],
+  videoDesc: string,
+  log?: AsrLogger,
+) {
+  const base64Audio = Buffer.from(audio).toString("base64");
+  const client = getQwenClient();
+  const model = process.env.QWEN_OMNI_MODEL || "qwen3.5-omni-plus";
+  const frameContentParts = frames.map((frame, index) => ({
+    type: "image_url" as const,
+    image_url: {
+      url: `data:${frame.contentType};base64,${frame.buffer.toString("base64")}`,
+    },
+    index,
+  }));
+
+  log?.(
+    `调用 Qwen Omni: model=${model}, audio=${Math.round(audio.length / 1024)}KB, frames=${frames.length}`,
+  );
+
+  const completion = await client.chat.completions.create({
+    model,
+    messages: [
+      {
+        role: "user",
+        content: [
+          ...frameContentParts.map(({ type, image_url }) => ({
+            type,
+            image_url,
+          })),
+          {
+            type: "input_audio",
+            input_audio: { data: `data:;base64,${base64Audio}`, format: "mp3" },
+          },
+          {
+            type: "text",
+            text: `${prompt}\n\n视频简介：\n${videoDesc || "无"}\n\n已附带视频三帧，时间点约为：${
+              frames
+                .map((frame) => `${(frame.timestampMs / 1000).toFixed(1)}s`)
+                .join("、") || "无"
+            }。请结合画面和音频判断内容标签。`,
+          },
+        ],
+      },
+    ],
+    modalities: ["text"],
+    stream: true,
+    stream_options: { include_usage: true },
+  });
+
+  let content = "";
+  for await (const chunk of completion) {
+    const deltaContent = chunk.choices?.[0]?.delta?.content;
+    if (typeof deltaContent === "string") content += deltaContent;
+  }
+
+  log?.(`Qwen Omni 原始输出: ${content}`);
+
+  if (!content.trim()) {
+    throw new Error("No ASR result returned from Qwen Omni");
+  }
+
+  try {
+    return parseAsrResult(content);
+  } catch {
+    throw new Error("Failed to parse ASR result JSON");
+  }
+}
+
+export async function transcriptAweme(
+  awemeId: string,
+  options: TranscriptAwemeOptions = {},
+): Promise<AsrResult> {
+  const log = options.log;
+  log?.(`查询视频: ${awemeId}`);
+  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;
+  log?.(`下载视频文件: ${vPath}`);
+  const buffer = await downloadFile(vPath);
+  log?.(`视频下载完成: ${Math.round(buffer.length / 1024)}KB`);
+
+  log?.("提取视频三帧");
+  const frames = await extractVideoFrames(buffer, {
+    count: 3,
+    durationMs: aweme.duration_ms,
+  });
+  log?.(
+    `视频帧提取完成: ${frames
+      .map((frame) => `${(frame.timestampMs / 1000).toFixed(1)}s`)
+      .join("、")}`,
+  );
+
+  log?.("提取音频");
+  const audioDat = await extractAudio(buffer, {
+    format: "mp3",
+    bitrateKbps: 128,
+  });
+
+  if (!audioDat || !audioDat.buffer) {
+    throw new Error("Failed to extract audio from video");
+  }
+  log?.(`音频提取完成: ${Math.round(audioDat.buffer.length / 1024)}KB`);
+
+  // 调用 Qwen Omni 生成转写与固定分类标签
+  const result = await transcriptAudio(
+    audioDat.buffer,
+    frames,
+    aweme.desc,
+    log,
+  );
+  log?.(`解析完成: tags=${result.content_tags.join("、")}`);
+
+  // 将转写结果持久化到数据库
+  log?.("写入数据库 VideoTranscript");
+  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,
+      content_tags: result.content_tags,
+    },
+    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,
+      content_tags: result.content_tags,
+    },
+  });
+  log?.("数据库写入完成");
+
+  return result;
+}

app/api/asr/prompt.md

@@ -0,0 +1,39 @@
+你将接收一段音频、视频中固定抽取的三帧画面，并可能收到一段视频描述作为辅助信息。请完成：语音活动检测（VAD）、声源分类、语音转文字，以及内容标签分类。
+
+只能输出一个合法 JSON 对象，不要输出 Markdown、解释文字或代码块。
+
+`content_tags` 必须从以下固定标签中选择 1 到 3 个，不能自造标签：
+
+美女、游戏、时政、美食、旅行、体育、科技、教育、宠物、音乐、影视、财经、生活、其他
+
+**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,
+  "content_tags": ["科技", "教育"]
+}
+```
+
+**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。",
+  "content_tags": ["音乐"]
+}
+```
+
+** 规则补充 **
+
+* 允许`language` 为多标签（如 "zh-CN,en"）或为 `null`（无发言时）。
+* `transcript` 按自然语义分句，每段尽量保持完整短句。
+* 必须结合三帧画面、音频内容和辅助视频描述判断标签；信息不足时使用 `其他`。

app/api/asr/route.ts

@@ -0,0 +1,22 @@
+export const runtime = "nodejs";
+
+import { type NextRequest, NextResponse } from "next/server";
+import { transcriptAweme } from ".";
+
+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 },
+    );
+  }
+  try {
+    const result = await transcriptAweme(awemeId, /* {log: (msg) => console.log(`[ASR] ${msg}`)} */);
+    return NextResponse.json(result);
+  } catch (error) {
+    console.error("ASR failed:", error);
+    return NextResponse.json({ error: "ASR failed" }, { status: 500 });
+  }
+}

app/api/feed/route.ts

@@ -1,7 +1,8 @@
-import { NextRequest, NextResponse } from "next/server";
+import { type NextRequest, NextResponse } from "next/server";
-import { prisma } from "@/lib/prisma";
 import type { FeedItem, FeedResponse } from "@/app/types/feed";
+import { isAsrContentTag } from "@/lib/asr-tags";
 import { getFileUrl } from "@/lib/minio";
+import { prisma } from "@/lib/prisma";
 
 // Contract
 // Inputs: search params { before?: ISOString, limit?: number }
@@ -11,24 +12,44 @@ export async function GET(req: NextRequest) {
   const { searchParams } = new URL(req.url);
   const limitParam = searchParams.get("limit");
   const beforeParam = searchParams.get("before");
+  const rawTag = (searchParams.get("tag") || "").trim();
+  const selectedTag = rawTag && isAsrContentTag(rawTag) ? rawTag : null;
+
+  if (rawTag && !selectedTag) {
+    return NextResponse.json({ error: "Invalid tag" }, { status: 400 });
+  }
 
   const limit = Math.min(Math.max(Number(limitParam ?? "24"), 1), 60); // 1..60
   const before = beforeParam ? new Date(beforeParam) : null;
+  const videoWhere = {
+    ...(before ? { created_at: { lt: before } } : {}),
+    ...(selectedTag
+      ? { transcript: { is: { content_tags: { has: selectedTag } } } }
+      : {}),
+  };
 
   // fetch chunk from both tables
   const [videos, posts] = await Promise.all([
     prisma.video.findMany({
-      where: before ? { created_at: { lt: before } } : undefined,
+      where: videoWhere,
       orderBy: { created_at: "desc" },
       take: limit,
-      include: { author: true },
+      include: {
-    }),
+        author: true,
-    prisma.imagePost.findMany({
+        transcript: { select: { content_tags: true } },
-      where: before ? { created_at: { lt: before } } : undefined,
+      },
-      orderBy: { created_at: "desc" },
-      take: limit,
-      include: { author: true, images: { orderBy: { order: "asc" }, take: 1 } },
     }),
+    selectedTag
+      ? Promise.resolve([])
+      : prisma.imagePost.findMany({
+          where: before ? { created_at: { lt: before } } : undefined,
+          orderBy: { created_at: "desc" },
+          take: limit,
+          include: {
+            author: true,
+            images: { orderBy: { order: "asc" }, take: 1 },
+          },
+        }),
   ]);
 
   const merged: FeedItem[] = [
@@ -47,29 +68,33 @@ export async function GET(req: NextRequest) {
         sec_uid: v.author.sec_uid,
       },
       likes: Number(v.digg_count),
+      content_tags: v.transcript?.content_tags ?? [],
     })),
-    ...posts.map((p) => ({
+    ...posts.map((p) => {
-      type: "image" as const,
+      const coverPath = p.images?.[0]?.url;
-      aweme_id: p.aweme_id,
+      return {
-      created_at: p.created_at,
+        type: "image" as const,
-      desc: p.desc,
+        aweme_id: p.aweme_id,
-      cover_url: getFileUrl(p.images?.[0]?.url ?? null),
+        created_at: p.created_at,
-      width: p.images?.[0]?.width ?? null,
+        desc: p.desc,
-      height: p.images?.[0]?.height ?? null,
+        cover_url: coverPath ? getFileUrl(coverPath) : null,
-      author: {
+        width: p.images?.[0]?.width ?? null,
-        nickname: p.author.nickname,
+        height: p.images?.[0]?.height ?? null,
-        avatar_url: getFileUrl(p.author.avatar_url ?? ""),
+        author: {
-        sec_uid: p.author.sec_uid,
+          nickname: p.author.nickname,
-      },
+          avatar_url: getFileUrl(p.author.avatar_url ?? ""),
-      likes: Number(p.digg_count),
+          sec_uid: p.author.sec_uid,
-    })),
+        },
+        likes: Number(p.digg_count),
+      };
+    }),
   ]
     .sort((a, b) => +new Date(b.created_at) - +new Date(a.created_at))
     .slice(0, limit);
 
   const nextCursor =
     merged.length > 0
-      ? new Date(merged[merged.length - 1].created_at as any).toISOString()
+      ? new Date(merged[merged.length - 1].created_at).toISOString()
       : null;
   const payload: FeedResponse = { items: merged, nextCursor };
   return NextResponse.json(payload);

app/api/fetcher/index.ts

@@ -21,7 +21,7 @@ import {
   releaseIsolatedContext,
 } from "@/app/api/fetcher/browser";
 import { extractFirstFrame } from "@/app/api/media";
-import { transcriptAweme } from "../stt";
+import { transcriptAweme } from "../asr";
 
 const DETAIL_PATH = "/aweme/v1/web/aweme/detail/";
 const COMMENT_PATH = "/aweme/v1/web/comment/list/";

app/api/media.ts

@@ -1,11 +1,28 @@
-import { execFile } from "child_process";
+import { execFile } from "node:child_process";
-import { promises as fs } from "fs";
+import { promises as fs } from "node:fs";
-import os from "os";
+import os from "node:os";
-import path from "path";
+import path from "node:path";
-import { promisify } from "util";
+import { promisify } from "node:util";
 
 const execFileAsync = promisify(execFile);
 
+function getErrorMessage(error: unknown) {
+  return error instanceof Error ? error.message : String(error);
+}
+
+function getErrorCode(error: unknown) {
+  return typeof error === "object" && error !== null && "code" in error
+    ? String((error as { code?: unknown }).code)
+    : null;
+}
+
+function isMissingExecutableError(error: unknown) {
+  return (
+    getErrorCode(error) === "ENOENT" ||
+    /not found|is not recognized/i.test(getErrorMessage(error))
+  );
+}
+
 /**
  * 使用 ffmpeg 从视频二进制中提取第一帧，返回 JPEG buffer
  */
@@ -40,18 +57,14 @@ export async function extractFirstFrame(
     await execFileAsync(ffmpegCmd, args, { windowsHide: true });
     const img = await fs.readFile(outPath);
     return { buffer: img, contentType: "image/jpeg", ext: "jpg" };
-  } catch (e: any) {
+  } catch (error) {
-    if (
+    if (isMissingExecutableError(error)) {
-      e &&
-      (e.code === "ENOENT" ||
-        /not found|is not recognized/i.test(String(e.message)))
-    ) {
       console.warn(
         "系统未检测到 ffmpeg，可安装并配置 PATH 或设置 FFMPEG_PATH 后启用封面提取。",
       );
       return null;
     }
-    throw e;
+    throw error;
   } finally {
     try {
       await fs.unlink(inPath);
@@ -62,6 +75,13 @@ export async function extractFirstFrame(
   }
 }
 
+export type ExtractedVideoFrame = {
+  buffer: Buffer;
+  contentType: string;
+  ext: string;
+  timestampMs: number;
+};
+
 /**
  * 使用 ffprobe 获取视频时长（毫秒）
  */
@@ -88,20 +108,16 @@ export async function getVideoDuration(
       windowsHide: true,
     });
     const durationSeconds = parseFloat(stdout.trim());
-    if (isNaN(durationSeconds)) return null;
+    if (Number.isNaN(durationSeconds)) return null;
     return Math.round(durationSeconds * 1000); // 转换为毫秒
-  } catch (e: any) {
+  } catch (error) {
-    if (
+    if (isMissingExecutableError(error)) {
-      e &&
-      (e.code === "ENOENT" ||
-        /not found|is not recognized/i.test(String(e.message)))
-    ) {
       console.warn(
         "系统未检测到 ffprobe，可安装并配置 PATH 或设置 FFPROBE_PATH 后启用时长提取。",
       );
       return null;
     }
-    console.warn(`获取视频时长失败: ${e?.message || e}`);
+    console.warn(`获取视频时长失败: ${getErrorMessage(error)}`);
     return null;
   } finally {
     try {
@@ -110,6 +126,102 @@ export async function getVideoDuration(
   }
 }
 
+function getFrameTimestamps(durationMs: number | null, count: number) {
+  if (!durationMs || durationMs <= 0) {
+    return Array.from({ length: count }, (_, index) => index * 1000);
+  }
+
+  if (count === 1) return [0];
+
+  const maxTimestamp = Math.max(0, durationMs - 200);
+  if (count === 3) {
+    return [
+      Math.min(maxTimestamp, Math.max(0, Math.round(durationMs * 0.1))),
+      Math.min(maxTimestamp, Math.max(0, Math.round(durationMs * 0.5))),
+      Math.min(maxTimestamp, Math.max(0, Math.round(durationMs * 0.9))),
+    ];
+  }
+
+  return Array.from({ length: count }, (_, index) => {
+    const ratio = count === 1 ? 0 : index / (count - 1);
+    return Math.min(maxTimestamp, Math.max(0, Math.round(durationMs * ratio)));
+  });
+}
+
+/**
+ * 使用 ffmpeg 从视频中提取固定数量的 JPEG 帧。
+ */
+export async function extractVideoFrames(
+  videoBuffer: Buffer,
+  opts?: { count?: number; durationMs?: number | null },
+): Promise<ExtractedVideoFrame[]> {
+  const ffmpegCmd = process.env.FFMPEG_PATH || "ffmpeg";
+  const tmpDir = os.tmpdir();
+  const base = `dy_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`;
+  const inPath = path.join(tmpDir, `${base}.mp4`);
+  const count = Math.max(1, opts?.count ?? 3);
+  const outPaths = Array.from({ length: count }, (_, index) =>
+    path.join(tmpDir, `${base}_${index}.jpg`),
+  );
+
+  try {
+    await fs.writeFile(inPath, videoBuffer);
+    const durationMs =
+      opts?.durationMs ?? (await getVideoDuration(videoBuffer));
+    const timestamps = getFrameTimestamps(durationMs, count);
+
+    const frames: ExtractedVideoFrame[] = [];
+    for (let index = 0; index < count; index++) {
+      const timestampSeconds = Math.max(0, timestamps[index] / 1000).toFixed(3);
+      const args = [
+        "-hide_banner",
+        "-loglevel",
+        "error",
+        "-ss",
+        timestampSeconds,
+        "-i",
+        inPath,
+        "-frames:v",
+        "1",
+        "-q:v",
+        "3",
+        "-vf",
+        "scale=768:-2:force_original_aspect_ratio=decrease",
+        "-f",
+        "image2",
+        "-y",
+        outPaths[index],
+      ];
+      await execFileAsync(ffmpegCmd, args, { windowsHide: true });
+      frames.push({
+        buffer: await fs.readFile(outPaths[index]),
+        contentType: "image/jpeg",
+        ext: "jpg",
+        timestampMs: timestamps[index],
+      });
+    }
+
+    return frames;
+  } catch (error) {
+    if (isMissingExecutableError(error)) {
+      console.warn(
+        "系统未检测到 ffmpeg，可安装并配置 PATH 或设置 FFMPEG_PATH 后启用视频帧提取。",
+      );
+      return [];
+    }
+    throw error;
+  } finally {
+    try {
+      await fs.unlink(inPath);
+    } catch {}
+    for (const outPath of outPaths) {
+      try {
+        await fs.unlink(outPath);
+      } catch {}
+    }
+  }
+}
+
 /**
  * 使用 ffmpeg 将视频二进制转换为音频，返回音频 Buffer
  *
@@ -189,20 +301,16 @@ export async function extractAudio(
     await execFileAsync(ffmpegCmd, args, { windowsHide: true });
     const audio = await fs.readFile(outPath);
     return { buffer: audio, contentType, ext };
-  } catch (e: any) {
+  } catch (error) {
-    if (
+    if (isMissingExecutableError(error)) {
-      e &&
-      (e.code === "ENOENT" ||
-        /not found|is not recognized/i.test(String(e.message)))
-    ) {
       console.warn(
         "系统未检测到 ffmpeg，可安装并配置 PATH 或设置 FFMPEG_PATH 后启用音频提取。",
       );
       return null;
     }
     // 一些环境可能缺少特定编码器（如 libmp3lame），提示并抛出原始错误
-    console.warn(`提取音频失败: ${e?.message || e}`);
+    console.warn(`提取音频失败: ${getErrorMessage(error)}`);
-    throw e;
+    throw error;
   } finally {
     try {
       await fs.unlink(inPath);

app/api/search/route.ts

@@ -1,11 +1,32 @@
+import { NextResponse } from "next/server";
+import { isAsrContentTag } from "@/lib/asr-tags";
 import { json } from "@/lib/json";
 import { getFileUrl } from "@/lib/minio";
-import { prisma } from "@/lib/prisma"; // 你的 Prisma 客户端实例
+import { prisma } from "@/lib/prisma";
-import { NextResponse } from "next/server";
+
+type SearchRow = {
+  id: string;
+  awemeId: string;
+  type: "video" | "image";
+  rank: number;
+  snippet: string;
+  contentTags: string[];
+};
+
+function sanitizeSnippetHtml(snippet: string) {
+  return snippet
+    .replaceAll("&", "&")
+    .replaceAll("<", "&lt;")
+    .replaceAll(">", "&gt;")
+    .replaceAll("&lt;mark&gt;", "<mark>")
+    .replaceAll("&lt;/mark&gt;", "</mark>");
+}
 
 export async function GET(req: Request) {
   const { searchParams } = new URL(req.url);
   const q = (searchParams.get("q") || "").trim();
+  const rawTag = (searchParams.get("tag") || "").trim();
+  const contentTag = rawTag && isAsrContentTag(rawTag) ? rawTag : null;
   const page = Math.max(1, Number(searchParams.get("page") || 1));
   const limit = Math.min(
     50,
@@ -13,7 +34,11 @@ export async function GET(req: Request) {
   );
   const offset = (page - 1) * limit;
 
-  if (!q) {
+  if (rawTag && !contentTag) {
+    return NextResponse.json({ error: "Invalid tag" }, { status: 400 });
+  }
+
+  if (!q && !contentTag) {
     return NextResponse.json({ results: [], total: 0 });
   }
 
@@ -23,90 +48,117 @@ export async function GET(req: Request) {
     // - 合并搜索来源：VideoTranscript.transcript_zh_tsv、Video.desc 与 ImagePost.desc
     // - 使用 UNION ALL 合并视频和图文的搜索结果
     // - rank 取两者匹配得分的最大值，snippet 优先展示转写匹配，否则展示 desc 匹配
-    const rows = await prisma.$queryRaw<
+    const rows = q
-      {
+      ? await prisma.$queryRaw<SearchRow[]>`
-        id: string;
+          WITH tsq AS (
-        awemeId: string;
+            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
-        type: "video" | "image";
+          )
-        rank: number;
+          -- 视频搜索结果
-        snippet: string;
+          SELECT
-      }[]
+            COALESCE(vt.id, 'desc:' || v.aweme_id) AS id,
-    >`
+            v.aweme_id AS "awemeId",
-      WITH tsq AS (
+            'video'::text AS type,
-        SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
+            GREATEST(
-      )
+              COALESCE(ts_rank(vt.transcript_zh_tsv, tsq.query), 0),
-      -- 视频搜索结果
+              COALESCE(ts_rank(to_tsvector('zhcfg', v.desc), tsq.query), 0)
-      SELECT
+            ) AS rank,
-        COALESCE(vt.id, 'desc:' || v.aweme_id) AS id,
+            CASE
-        v.aweme_id AS "awemeId",
+              WHEN vt.transcript_zh_tsv @@ tsq.query THEN
-        'video'::text AS type,
+                ts_headline(
-        GREATEST(
+                  'zhcfg',
-          COALESCE(ts_rank(vt.transcript_zh_tsv, tsq.query), 0),
+                  array_to_string(vt.transcript, ' '),
-          COALESCE(ts_rank(to_tsvector('zhcfg', v.desc), tsq.query), 0)
+                  tsq.query,
-        ) AS rank,
+                  'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
-        CASE
+                )
-          WHEN vt.transcript_zh_tsv @@ tsq.query THEN
+              ELSE
+                ts_headline(
+                  'zhcfg',
+                  v.desc,
+                  tsq.query,
+                  'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
+                )
+            END AS snippet,
+            COALESCE(vt.content_tags, ARRAY[]::text[]) AS "contentTags"
+          FROM "Video" v
+          LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id,
+               tsq
+          WHERE
+            (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query)
+            AND CASE
+              WHEN ${contentTag}::text IS NULL THEN TRUE
+              ELSE vt.content_tags @> ARRAY[${contentTag}]::text[]
+            END
+
+          UNION ALL
+
+          -- 图文搜索结果
+          SELECT
+            'img_desc:' || ip.aweme_id AS id,
+            ip.aweme_id AS "awemeId",
+            'image'::text AS type,
+            ts_rank(to_tsvector('zhcfg', ip.desc), tsq.query) AS rank,
             ts_headline(
               'zhcfg',
-              array_to_string(vt.transcript, ' '),
+              ip.desc,
               tsq.query,
               'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
-            )
+            ) AS snippet,
-          ELSE
+            ARRAY[]::text[] AS "contentTags"
-            ts_headline(
+          FROM "ImagePost" ip, tsq
-              'zhcfg',
+          WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
+            AND ${contentTag}::text IS NULL
+
+          ORDER BY rank DESC
+          OFFSET ${offset}
+          LIMIT ${limit};
+        `
+      : await prisma.$queryRaw<SearchRow[]>`
+          SELECT
+            vt.id AS id,
+            v.aweme_id AS "awemeId",
+            'video'::text AS type,
+            1.0::double precision AS rank,
+            COALESCE(
+              NULLIF(array_to_string(vt.transcript[1:2], ' '), ''),
+              vt.non_speech_summary,
               v.desc,
-              tsq.query,
+              '标签匹配'
-              'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
+            ) AS snippet,
-            )
+            vt.content_tags AS "contentTags"
-        END AS snippet
+          FROM "VideoTranscript" vt
-      FROM "Video" v
+          JOIN "Video" v ON v.aweme_id = vt."videoId"
-      LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id,
+          WHERE vt.content_tags @> ARRAY[${contentTag}]::text[]
-           tsq
+          ORDER BY v.created_at DESC
-      WHERE
+          OFFSET ${offset}
-        (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query)
+          LIMIT ${limit};
-      
+        `;
-      UNION ALL
-      
-      -- 图文搜索结果
-      SELECT
-        'img_desc:' || ip.aweme_id AS id,
-        ip.aweme_id AS "awemeId",
-        'image'::text AS type,
-        ts_rank(to_tsvector('zhcfg', ip.desc), tsq.query) AS rank,
-        ts_headline(
-          'zhcfg',
-          ip.desc,
-          tsq.query,
-          'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
-        ) AS snippet
-      FROM "ImagePost" ip, tsq
-      WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
-      
-      ORDER BY rank DESC
-      OFFSET ${offset}
-      LIMIT ${limit};
-    `;
 
     // 查询总数（参数化，避免注入）
-    const totalRows = await prisma.$queryRaw<
+    const totalRows = q
-      {
+      ? await prisma.$queryRaw<{ count: number }[]>`
-        count: number;
+          WITH tsq AS (
-      }[]
+            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
-    >`
+          )
-      WITH tsq AS (
+          SELECT (
-        SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
+            (SELECT COUNT(*)::int FROM "Video" v
-      )
+             LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
-      SELECT (
+             WHERE (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query)
-        (SELECT COUNT(*)::int FROM "Video" v
+               AND CASE
-         LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
+                 WHEN ${contentTag}::text IS NULL THEN TRUE
-         WHERE (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query))
+                 ELSE vt.content_tags @> ARRAY[${contentTag}]::text[]
-        +
+               END)
-        (SELECT COUNT(*)::int FROM "ImagePost" ip
+            +
-         WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query)
+            (SELECT COUNT(*)::int FROM "ImagePost" ip
-      ) AS count
+             WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
-      FROM tsq;
+               AND ${contentTag}::text IS NULL)
-    `;
+          ) AS count
+          FROM tsq;
+        `
+      : await prisma.$queryRaw<{ count: number }[]>`
+          SELECT COUNT(*)::int AS count
+          FROM "VideoTranscript" vt
+          WHERE vt.content_tags @> ARRAY[${contentTag}]::text[];
+        `;
 
     // 分离视频和图文ID
     const videoIds = rows
@@ -176,6 +228,7 @@ export async function GET(req: Request) {
     return json({
       results: rows.map((r) => ({
         ...r,
+        snippet: sanitizeSnippetHtml(r.snippet),
         video:
           r.type === "video"
             ? videos.find((v) => v.aweme_id === r.awemeId)

app/api/stt/index.ts

@@ -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;
-}

app/api/stt/prompt.md

@@ -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）。

app/api/stt/route.ts

@@ -1,23 +0,0 @@
-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 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 script = await transcriptAweme(awemeId);
-  return NextResponse.json(script);
-}

app/aweme/[awemeId]/components/TranscriptPanel.tsx

@@ -1,6 +1,6 @@
 "use client";
 
-import { X, Copy, CheckCheck, Check } from "lucide-react";
+import { Check, CheckCheck, Copy, X } from "lucide-react";
 import { useState } from "react";
 import type { VideoTranscript } from "../types";
 
@@ -38,18 +38,23 @@ export function TranscriptPanel({
 
   if (!transcript?.speech_detected || !transcript.transcript?.length) {
     return (
-      <div
+      <div className="fixed inset-0 z-50 flex items-center justify-center bg-black/60 backdrop-blur-sm">
-        className="fixed inset-0 z-50 flex items-center justify-center bg-black/60 backdrop-blur-sm"
+        <button
-        onClick={onClose}
+          type="button"
-      >
+          aria-label="关闭"
+          className="absolute inset-0 h-full w-full cursor-default"
+          onClick={onClose}
+        />
         <div
-          className="relative w-full max-w-2xl max-h-[80vh] m-4 bg-zinc-900/95 backdrop-blur-xl border border-white/10 rounded-2xl shadow-2xl"
+          role="dialog"
-          onClick={(e) => e.stopPropagation()}
+          aria-modal="true"
+          className="relative z-10 w-full max-w-2xl max-h-[80vh] m-4 bg-zinc-900/95 backdrop-blur-xl border border-white/10 rounded-2xl shadow-2xl"
         >
           {/* 头部 */}
           <div className="flex items-center justify-between p-4 border-b border-white/10">
             <h2 className="text-lg font-semibold text-white">语音转录</h2>
             <button
+              type="button"
               onClick={onClose}
               className="p-2 rounded-full hover:bg-white/10 transition-colors"
               aria-label="关闭"
@@ -68,13 +73,17 @@ export function TranscriptPanel({
   }
 
   return (
-    <div
+    <div className="fixed inset-0 z-50 flex items-center justify-center bg-black/60 backdrop-blur-sm">
-      className="fixed inset-0 z-50 flex items-center justify-center bg-black/60 backdrop-blur-sm"
+      <button
-      onClick={onClose}
+        type="button"
-    >
+        aria-label="关闭"
+        className="absolute inset-0 h-full w-full cursor-default"
+        onClick={onClose}
+      />
       <div
-        className="relative w-full max-w-2xl max-h-[80vh] m-4 bg-zinc-900/95 backdrop-blur-xl border border-white/10 rounded-2xl shadow-2xl flex flex-col"
+        role="dialog"
-        onClick={(e) => e.stopPropagation()}
+        aria-modal="true"
+        className="relative z-10 w-full max-w-2xl max-h-[80vh] m-4 bg-zinc-900/95 backdrop-blur-xl border border-white/10 rounded-2xl shadow-2xl flex flex-col"
       >
         {/* 头部 */}
         <div className="flex items-center justify-between p-4 border-b border-white/10 flex-shrink-0">
@@ -85,9 +94,22 @@ export function TranscriptPanel({
                 语言: {transcript.language}
               </p>
             )}
+            {transcript.content_tags?.length > 0 && (
+              <div className="mt-2 flex flex-wrap gap-1.5">
+                {transcript.content_tags.map((tag) => (
+                  <span
+                    key={tag}
+                    className="rounded bg-blue-500/15 px-2 py-0.5 text-xs text-blue-200"
+                  >
+                    {tag}
+                  </span>
+                ))}
+              </div>
+            )}
           </div>
           <div className="flex items-center gap-2">
             <button
+              type="button"
               onClick={handleCopyAll}
               className="flex items-center gap-2 px-3 py-2 text-sm bg-blue-600/20 text-blue-300 rounded-lg hover:bg-blue-600/30 transition-colors"
             >
@@ -104,6 +126,7 @@ export function TranscriptPanel({
               )}
             </button>
             <button
+              type="button"
               onClick={onClose}
               className="p-2 rounded-full hover:bg-white/10 transition-colors"
               aria-label="关闭"
@@ -117,9 +140,8 @@ export function TranscriptPanel({
         <div className="flex-1 overflow-y-auto p-4 space-y-2">
           {transcript.transcript.map((text, index) => (
             <div
-              key={index}
+              key={`${index}-${text}`}
-              className="group bg-white/5 rounded-lg p-3 hover:bg-white/10 transition-colors cursor-pointer"
+              className="group bg-white/5 rounded-lg p-3 hover:bg-white/10 transition-colors"
-              onClick={() => handleCopy(text, index)}
             >
               <div className="flex items-start justify-between gap-3">
                 <div className="flex-1 min-w-0">
@@ -133,8 +155,9 @@ export function TranscriptPanel({
                   </p>
                 </div>
                 <button
+                  type="button"
                   onClick={() => handleCopy(text, index)}
-                  className="flex-shrink-0 p-2 rounded-lg opacity-0 group-hover:opacity-100 hover:bg-white/10 transition-all"
+                  className="flex-shrink-0 p-2 rounded-lg opacity-0 group-hover:opacity-100 focus:opacity-100 hover:bg-white/10 transition-all"
                   aria-label="复制"
                 >
                   {copiedIndex === index ? (

app/aweme/[awemeId]/types.ts

@@ -56,6 +56,7 @@ export type VideoTranscript = {
   speech_detected: boolean;
   language: string | null;
   transcript: string[];
+  content_tags: string[];
 };
 
 export type Neighbors = {

app/components/FeedMasonry.tsx

@@ -1,9 +1,16 @@
 "use client";
-import { useCallback, useEffect, useMemo, useRef, useState } from "react";
-import Link from "next/link";
-import HoverVideo from "./HoverVideo";
 import { ThumbsUp } from "lucide-react";
+import Image from "next/image";
+import Link from "next/link";
+import { useCallback, useEffect, useRef, useState } from "react";
 import type { FeedItem, FeedResponse } from "@/app/types/feed";
+import HoverVideo from "./HoverVideo";
+
+const COLUMN_KEYS = ["feed-col-0", "feed-col-1", "feed-col-2", "feed-col-3"];
+
+function getAspectRatio(item: FeedItem) {
+  return item.width && item.height ? `${item.width}/${item.height}` : undefined;
+}
 
 type Props = {
   initialItems: FeedItem[];
@@ -58,7 +65,7 @@ export default function FeedMasonry({
       mediaH = (9 / 16) * colWidth; // 常见视频比例
     }
     // 文本 + 作者栏的高度粗估
-    const textH = 48; // 标题+标签区域
+    const textH = item.content_tags?.length ? 72 : 48; // 标题+标签区域
     const authorH = 48; // 作者行
     const gap = 16; // 卡片内边距/间隙
     return mediaH + textH + authorH + gap;
@@ -92,8 +99,7 @@ export default function FeedMasonry({
     }
     setColumns(newCols);
     setColHeights(newHeights);
-    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [columnCount, estimateItemHeight, initialItems]);
-  }, [columnCount]);
 
   const fetchMore = useCallback(async () => {
     if (loading || ended) return;
@@ -136,7 +142,15 @@ export default function FeedMasonry({
     } finally {
       setLoading(false);
     }
-  }, [cursor, ended, loading, columnCount, colHeights, estimateItemHeight]);
+  }, [
+    cursor,
+    ended,
+    loading,
+    columnCount,
+    colHeights,
+    estimateItemHeight,
+    fetchUrl,
+  ]);
 
   useEffect(() => {
     const el = sentinelRef.current;
@@ -168,29 +182,43 @@ export default function FeedMasonry({
           <div
             className="relative w-full"
             style={{
-              aspectRatio:
+              aspectRatio: getAspectRatio(item),
-                `${item.width && item.height ? `${item.width}/${item.height}` : ""}` as any,
             }}
           >
             {item.type === "video" ? (
               <HoverVideo
-                videoUrl={(item as any).video_url}
+                videoUrl={item.video_url}
                 coverUrl={item.cover_url}
                 className="absolute inset-0 w-full h-full"
               />
             ) : (
-              <img
+              <Image
-                loading="lazy"
+                fill
                 src={item.cover_url || "/placeholder.svg"}
                 alt={item.desc?.slice(0, 20) || "image"}
+                unoptimized
                 className="absolute inset-0 w-full h-full object-cover"
               />
             )}
             <div className="pointer-events-none absolute inset-0 bg-gradient-to-t from-black/50 via-black/0 to-black/0 opacity-70" />
             <div className="absolute left-3 bottom-3 right-3 flex items-end justify-between gap-3">
-              <p className="text-white/95 text-sm leading-tight line-clamp-2 drop-shadow">
+              <div className="min-w-0">
-                {item.desc}
+                <p className="text-white/95 text-sm leading-tight line-clamp-2 drop-shadow">
-              </p>
+                  {item.desc}
+                </p>
+                {!!item.content_tags?.length && (
+                  <div className="mt-2 flex flex-wrap gap-1">
+                    {item.content_tags.map((tag) => (
+                      <span
+                        key={tag}
+                        className="rounded-full bg-blue-500/80 px-2 py-0.5 text-xs text-white shadow"
+                      >
+                        {tag}
+                      </span>
+                    ))}
+                  </div>
+                )}
+              </div>
               <span className="shrink-0 inline-flex items-center gap-2 rounded-full bg-white/85 px-2 py-1 text-xs text-zinc-800">
                 {item.type === "video" ? "视频" : "图文"}
               </span>
@@ -206,9 +234,12 @@ export default function FeedMasonry({
             >
               <div className="size-6 rounded-full overflow-hidden bg-zinc-200 shrink-0">
                 {item.author.avatar_url ? (
-                  <img
+                  <Image
                     src={item.author.avatar_url}
                     alt="avatar"
+                    width={24}
+                    height={24}
+                    unoptimized
                     className="w-full h-full object-cover"
                   />
                 ) : null}
@@ -221,9 +252,12 @@ export default function FeedMasonry({
             <div className="flex items-center gap-2 min-w-0 flex-1">
               <div className="size-6 rounded-full overflow-hidden bg-zinc-200 shrink-0">
                 {item.author.avatar_url ? (
-                  <img
+                  <Image
                     src={item.author.avatar_url}
                     alt="avatar"
+                    width={24}
+                    height={24}
+                    unoptimized
                     className="w-full h-full object-cover"
                   />
                 ) : null}
@@ -254,7 +288,7 @@ export default function FeedMasonry({
         }}
       >
         {columns.map((col, idx) => (
-          <div key={idx} className="flex flex-col">
+          <div key={COLUMN_KEYS[idx]} className="flex flex-col">
             {col.map((item) => renderCard(item))}
           </div>
         ))}

app/components/SearchBox.tsx

@@ -1,10 +1,10 @@
 "use client";
 
-import { useState } from "react";
-import { useRouter } from "next/navigation";
 import { Search } from "lucide-react";
+import { useRouter } from "next/navigation";
+import { useState } from "react";
 
-export default function SearchBox() {
+export default function SearchBox({ tag }: { tag?: string | null }) {
   const [query, setQuery] = useState("");
   const router = useRouter();
 
@@ -12,7 +12,9 @@ export default function SearchBox() {
     e.preventDefault();
     const trimmed = query.trim();
     if (!trimmed) return;
-    router.push(`/search?q=${encodeURIComponent(trimmed)}`);
+    const params = new URLSearchParams({ q: trimmed });
+    if (tag) params.set("tag", tag);
+    router.push(`/search?${params.toString()}`);
   };
 
   return (

app/page.tsx

@@ -1,9 +1,11 @@
+import type { Metadata } from "next";
+import Link from "next/link";
+import { ASR_CONTENT_TAGS, isAsrContentTag } from "@/lib/asr-tags";
+import { getFileUrl } from "@/lib/minio";
 import { prisma } from "@/lib/prisma";
 import FeedMasonry from "./components/FeedMasonry";
 import SearchBox from "./components/SearchBox";
 import type { FeedItem } from "./types/feed";
-import type { Metadata } from "next";
-import { getFileUrl } from "@/lib/minio";
 
 export const dynamic = "force-dynamic";
 
@@ -12,18 +14,37 @@ export const metadata: Metadata = {
   description: "抖歪作品集，记录当下时代的精彩瞬间",
 };
 
-export default async function Home() {
+export default async function Home({
+  searchParams,
+}: {
+  searchParams?: { tag?: string };
+}) {
+  const requestedTag = searchParams?.tag?.trim() || "";
+  const selectedTag =
+    requestedTag && isAsrContentTag(requestedTag) ? requestedTag : null;
+
   const [videos, posts] = await Promise.all([
     prisma.video.findMany({
+      where: selectedTag
+        ? { transcript: { is: { content_tags: { has: selectedTag } } } }
+        : undefined,
       orderBy: { created_at: "desc" },
       take: 60,
-      include: { author: true },
+      include: {
-    }),
+        author: true,
-    prisma.imagePost.findMany({
+        transcript: { select: { content_tags: true } },
-      orderBy: { created_at: "desc" },
+      },
-      take: 60,
-      include: { author: true, images: { orderBy: { order: "asc" }, take: 1 } },
     }),
+    selectedTag
+      ? Promise.resolve([])
+      : prisma.imagePost.findMany({
+          orderBy: { created_at: "desc" },
+          take: 60,
+          include: {
+            author: true,
+            images: { orderBy: { order: "asc" }, take: 1 },
+          },
+        }),
   ]);
 
   const feed: FeedItem[] = [
@@ -42,22 +63,26 @@ export default async function Home() {
         sec_uid: v.author.sec_uid,
       },
       likes: Number(v.digg_count),
+      content_tags: v.transcript?.content_tags ?? [],
     })),
-    ...posts.map((p) => ({
+    ...posts.map((p) => {
-      type: "image" as const,
+      const coverPath = p.images?.[0]?.url;
-      aweme_id: p.aweme_id,
+      return {
-      created_at: p.created_at,
+        type: "image" as const,
-      desc: p.desc,
+        aweme_id: p.aweme_id,
-      cover_url: getFileUrl(p.images?.[0]?.url ?? null),
+        created_at: p.created_at,
-      width: p.images?.[0]?.width ?? null,
+        desc: p.desc,
-      height: p.images?.[0]?.height ?? null,
+        cover_url: coverPath ? getFileUrl(coverPath) : null,
-      author: {
+        width: p.images?.[0]?.width ?? null,
-        nickname: p.author.nickname,
+        height: p.images?.[0]?.height ?? null,
-        avatar_url: getFileUrl(p.author.avatar_url ?? ""),
+        author: {
-        sec_uid: p.author.sec_uid,
+          nickname: p.author.nickname,
-      },
+          avatar_url: getFileUrl(p.author.avatar_url ?? ""),
-      likes: Number(p.digg_count),
+          sec_uid: p.author.sec_uid,
-    })),
+        },
+        likes: Number(p.digg_count),
+      };
+    }),
   ]
     //.sort(() => Math.random() - 0.5)
     .sort((a, b) => +new Date(b.created_at) - +new Date(a.created_at));
@@ -75,20 +100,53 @@ export default async function Home() {
         </h1>
 
         <div className="w-full max-w-xl mb-12">
-          <SearchBox />
+          <SearchBox tag={selectedTag} />
+        </div>
+
+        <div className="mb-8 flex w-full max-w-4xl flex-wrap justify-center gap-2">
+          <Link
+            href="/"
+            className={`rounded-full border px-3 py-1.5 text-sm font-medium transition-colors ${
+              selectedTag
+                ? "border-zinc-200 bg-white/70 text-zinc-600 hover:border-zinc-300 dark:border-white/10 dark:bg-white/5 dark:text-zinc-300"
+                : "border-blue-500 bg-blue-600 text-white"
+            }`}
+          >
+            全部
+          </Link>
+          {ASR_CONTENT_TAGS.map((tag) => (
+            <Link
+              key={tag}
+              href={`/?tag=${encodeURIComponent(tag)}`}
+              className={`rounded-full border px-3 py-1.5 text-sm font-medium transition-colors ${
+                selectedTag === tag
+                  ? "border-blue-500 bg-blue-600 text-white"
+                  : "border-zinc-200 bg-white/70 text-zinc-600 hover:border-zinc-300 dark:border-white/10 dark:bg-white/5 dark:text-zinc-300 dark:hover:text-white"
+              }`}
+            >
+              {tag}
+            </Link>
+          ))}
         </div>
 
         <div className="w-full">
           {(() => {
             const initial = feed.slice(0, 24);
-            const cursor =
+            const lastItem = initial.at(-1);
-              initial.length > 0
+            const cursor = lastItem
-                ? new Date(
+              ? new Date(lastItem.created_at).toISOString()
-                    initial[initial.length - 1].created_at as any,
+              : null;
-                  ).toISOString()
-                : null;
             return (
-              <FeedMasonry initialItems={initial} initialCursor={cursor} />
+              <FeedMasonry
+                key={selectedTag ?? "all"}
+                initialItems={initial}
+                initialCursor={cursor}
+                fetchUrl={
+                  selectedTag
+                    ? `/api/feed?tag=${encodeURIComponent(selectedTag)}`
+                    : "/api/feed"
+                }
+              />
             );
           })()}
         </div>

app/search/SearchClient.tsx

@@ -1,9 +1,11 @@
 "use client";
 
-import { FormEvent, useCallback, useEffect, useState } from "react";
+import { ArrowLeft, MessageSquare, Search, X } from "lucide-react";
-import { useSearchParams, useRouter } from "next/navigation";
+import Image from "next/image";
 import Link from "next/link";
-import { Search, ArrowLeft, X, MessageSquare } from "lucide-react";
+import { useRouter, useSearchParams } from "next/navigation";
+import { type FormEvent, useCallback, useEffect, useState } from "react";
+import { ASR_CONTENT_TAGS, isAsrContentTag } from "@/lib/asr-tags";
 
 // 匹配搜索结果类型（适配后端API）
 type SearchResultItem = {
@@ -12,6 +14,7 @@ type SearchResultItem = {
   type: "video" | "image";
   rank: number;
   snippet: string; // 后端返回的高亮片段，已包含<mark>标签
+  contentTags: string[];
   video?: {
     aweme_id: string;
     desc: string;
@@ -47,9 +50,20 @@ export default function SearchClient({
   const [results, setResults] = useState<SearchResultItem[]>([]);
   const [loading, setLoading] = useState(false);
   const [searched, setSearched] = useState(false);
+  const [selectedTag, setSelectedTag] = useState<string | null>(null);
 
-  const performSearch = useCallback(async (q: string) => {
+  const buildSearchHref = useCallback((q: string, tag: string | null) => {
-    if (!q.trim()) {
+    const params = new URLSearchParams();
+    const trimmed = q.trim();
+    if (trimmed) params.set("q", trimmed);
+    if (tag) params.set("tag", tag);
+    const queryString = params.toString();
+    return queryString ? `/search?${queryString}` : "/search";
+  }, []);
+
+  const performSearch = useCallback(async (q: string, tag: string | null) => {
+    const trimmed = q.trim();
+    if (!trimmed && !tag) {
       setResults([]);
       setSearched(false);
       return;
@@ -57,9 +71,10 @@ export default function SearchClient({
     setLoading(true);
     setSearched(true);
     try {
-      const res = await fetch(
+      const params = new URLSearchParams({ limit: "60" });
-        `/api/search?q=${encodeURIComponent(q.trim())}&limit=60`,
+      if (trimmed) params.set("q", trimmed);
-      );
+      if (tag) params.set("tag", tag);
+      const res = await fetch(`/api/search?${params.toString()}`);
       if (!res.ok) throw new Error("Search failed");
       const data = await res.json();
       setResults(data.results || []);
@@ -74,32 +89,64 @@ export default function SearchClient({
   // 初始 & URL 变化时同步（用 searchParams 订阅 URL 更稳妥）
   useEffect(() => {
     const q = searchParams.get("q") || "";
+    const tagParam = searchParams.get("tag") || "";
+    const tag = tagParam && isAsrContentTag(tagParam) ? tagParam : null;
     setQuery(q);
-    if (q) performSearch(q);
+    setSelectedTag(tag);
+    if (q || tag) {
+      performSearch(q, tag);
+    } else {
+      setResults([]);
+      setSearched(false);
+    }
   }, [searchParams, performSearch]);
 
   const handleSearch = (e: FormEvent) => {
     e.preventDefault();
     const trimmed = query.trim();
-    if (!trimmed) return;
+    if (!trimmed && !selectedTag) return;
-    router.push(`/search?q=${encodeURIComponent(trimmed)}`);
+    router.push(buildSearchHref(trimmed, selectedTag));
-    // 可省略：上面的 push 会触发上面的 effect 从 URL 拉取并搜索
+  };
-    // performSearch(trimmed);
+
+  const handleTagSelect = (tag: string) => {
+    const nextTag = selectedTag === tag ? null : tag;
+    setSelectedTag(nextTag);
+    router.push(buildSearchHref(query, nextTag));
   };
 
   const handleClear = () => {
     setQuery("");
+    setSelectedTag(null);
     setResults([]);
     setSearched(false);
     router.push("/search");
   };
 
+  const renderSnippet = (snippet: string) => {
+    let highlighted = false;
+    let offset = 0;
+    return snippet.split(/(<mark>|<\/mark>)/g).map((part) => {
+      offset += part.length;
+      if (part === "<mark>") {
+        highlighted = true;
+        return null;
+      }
+      if (part === "</mark>") {
+        highlighted = false;
+        return null;
+      }
+      if (!part) return null;
+      return highlighted ? <mark key={`${offset}-${part}`}>{part}</mark> : part;
+    });
+  };
+
   return (
     <div className="min-h-screen bg-zinc-950 text-white">
       {/* 顶部搜索栏 */}
       <header className="sticky top-0 z-50 bg-zinc-900/80 backdrop-blur-xl border-b border-white/10">
         <div className="max-w-7xl mx-auto px-4 py-4 flex items-center gap-4">
           <button
+            type="button"
             onClick={() => router.push("/")}
             className="p-2 rounded-full hover:bg-white/10 transition-colors"
             aria-label="返回首页"
@@ -118,13 +165,12 @@ export default function SearchClient({
                 onChange={(e) => setQuery(e.target.value)}
                 placeholder="搜索视频、图文描述或语音内容..."
                 className="w-full bg-white/5 border border-white/10 rounded-full px-5 py-3 pl-12 pr-12 text-white placeholder-white/40 focus:outline-none focus:ring-2 focus:ring-blue-500/50 focus:border-blue-500/50 transition-all"
-                autoFocus
               />
               <Search
                 size={20}
                 className="absolute left-4 top-1/2 -translate-y-1/2 text-white/40"
               />
-              {query && (
+              {(query || selectedTag) && (
                 <button
                   type="button"
                   onClick={handleClear}
@@ -138,12 +184,30 @@ export default function SearchClient({
             <button
               type="submit"
               className="px-6 py-3 bg-blue-600 hover:bg-blue-700 rounded-full font-medium transition-colors disabled:opacity-50 disabled:cursor-not-allowed"
-              disabled={loading || !query.trim()}
+              disabled={loading || (!query.trim() && !selectedTag)}
             >
               {loading ? "搜索中..." : "搜索"}
             </button>
           </form>
         </div>
+        <div className="max-w-7xl mx-auto px-4 pb-4">
+          <div className="flex gap-2 overflow-x-auto">
+            {ASR_CONTENT_TAGS.map((tag) => (
+              <button
+                key={tag}
+                type="button"
+                onClick={() => handleTagSelect(tag)}
+                className={`flex-shrink-0 rounded-full border px-3 py-1.5 text-xs font-medium transition-colors ${
+                  selectedTag === tag
+                    ? "border-blue-400 bg-blue-500/20 text-blue-200"
+                    : "border-white/10 bg-white/5 text-white/60 hover:border-white/20 hover:text-white"
+                }`}
+              >
+                {tag}
+              </button>
+            ))}
+          </div>
+        </div>
       </header>
 
       {/* 结果区域 */}
@@ -165,7 +229,7 @@ export default function SearchClient({
           </div>
         )}
 
-        {!loading && !searched && !query && (
+        {!loading && !searched && !query && !selectedTag && (
           <div className="flex flex-col items-center justify-center py-20 text-white/40">
             <Search size={64} className="mb-4 opacity-20" />
             <p className="text-xl">输入关键词开始搜索</p>
@@ -207,11 +271,18 @@ export default function SearchClient({
                         target="_blank"
                         className="flex-shrink-0 w-64 h-80 sm:w-70 sm:h-96 rounded-xl overflow-hidden bg-zinc-900 group"
                       >
-                        <img
+                        {content.cover_url ? (
-                          src={content.cover_url || ""}
+                          <Image
-                          alt={content.desc}
+                            src={content.cover_url}
-                          className="w-full h-full object-cover transition-transform duration-300 group-hover:scale-105"
+                            alt={content.desc}
-                        />
+                            width={280}
+                            height={384}
+                            unoptimized
+                            className="w-full h-full object-cover transition-transform duration-300 group-hover:scale-105"
+                          />
+                        ) : (
+                          <div className="h-full w-full bg-zinc-800" />
+                        )}
                       </Link>
 
                       {/* 右侧：信息区 */}
@@ -231,6 +302,14 @@ export default function SearchClient({
                             <MessageSquare size={12} />
                             匹配度: {(item.rank * 100).toFixed(1)}%
                           </span>
+                          {item.contentTags.map((tag) => (
+                            <span
+                              key={tag}
+                              className="px-2 py-0.5 text-xs rounded bg-blue-500/15 text-blue-200"
+                            >
+                              {tag}
+                            </span>
+                          ))}
                         </div>
 
                         {/* 描述 */}
@@ -245,11 +324,18 @@ export default function SearchClient({
 
                         {/* 作者信息 */}
                         <div className="flex items-center gap-2 mb-4">
-                          <img
+                          {content.author.avatar_url ? (
-                            src={content.author.avatar_url || ""}
+                            <Image
-                            alt={content.author.nickname}
+                              src={content.author.avatar_url}
-                            className="w-6 h-6 rounded-full ring-1 ring-white/20"
+                              alt={content.author.nickname}
-                          />
+                              width={24}
+                              height={24}
+                              unoptimized
+                              className="w-6 h-6 rounded-full ring-1 ring-white/20"
+                            />
+                          ) : (
+                            <div className="h-6 w-6 rounded-full bg-white/10 ring-1 ring-white/20" />
+                          )}
                           <span className="text-xs text-white/60">
                             {content.author.nickname}
                           </span>
@@ -264,11 +350,12 @@ export default function SearchClient({
                             </h4>
                             <div
                               className="text-xs text-white/70 bg-white/5 rounded-lg p-2 leading-relaxed"
-                              dangerouslySetInnerHTML={{ __html: item.snippet }}
                               style={{
                                 wordBreak: "break-word",
                               }}
-                            />
+                            >
+                              {renderSnippet(item.snippet)}
+                            </div>
                           </div>
                         </div>
                       </div>

app/types/feed.ts

@@ -15,6 +15,7 @@ export type FeedItem = (
   cover_url: string | null;
   width?: number | null;
   height?: number | null;
+  content_tags?: string[];
 };
 
 export interface FeedResponse {

lib/asr-tags.ts

@@ -0,0 +1,40 @@
+export const ASR_CONTENT_TAGS = [
+  "美女",
+  "游戏",
+  "时政",
+  "美食",
+  "旅行",
+  "体育",
+  "科技",
+  "教育",
+  "宠物",
+  "音乐",
+  "影视",
+  "财经",
+  "生活",
+  "其他",
+] as const;
+
+export type AsrContentTag = (typeof ASR_CONTENT_TAGS)[number];
+
+const asrContentTagSet = new Set<string>(ASR_CONTENT_TAGS);
+
+export function isAsrContentTag(value: string): value is AsrContentTag {
+  return asrContentTagSet.has(value);
+}
+
+export function normalizeAsrContentTags(tags: unknown): AsrContentTag[] {
+  const rawTags = Array.isArray(tags)
+    ? tags
+    : typeof tags === "string"
+      ? tags.split(/[、,，\s]+/).filter(Boolean)
+      : [];
+
+  const normalized = rawTags.filter(
+    (tag): tag is AsrContentTag =>
+      typeof tag === "string" && isAsrContentTag(tag),
+  );
+
+  const uniqueTags = [...new Set(normalized)].slice(0, 3);
+  return uniqueTags.length > 0 ? uniqueTags : ["其他"];
+}

package.json

@@ -6,6 +6,7 @@
     "dev": "next dev --turbopack",
     "build": "next build --turbopack",
     "start": "next start",
+    "asr": "node scripts/asr-cli.cjs",
     "lint": "biome check",
     "format": "biome format --write"
   },
@@ -21,7 +22,6 @@
     "puppeteer-extra-plugin-stealth": "^2.11.2",
     "react": "19.1.0",
     "react-dom": "19.1.0",
-    "undici": "^7.16.0",
     "zod": "^4.1.12"
   },
   "devDependencies": {

prisma/migrations/20260429_add_asr_content_tags/migration.sql

@@ -0,0 +1,5 @@
+ALTER TABLE "VideoTranscript"
+ADD COLUMN "content_tags" TEXT[] NOT NULL DEFAULT ARRAY[]::TEXT[];
+
+CREATE INDEX "video_transcript_content_tags_idx"
+ON "VideoTranscript" USING GIN ("content_tags");

prisma/schema.prisma

@@ -145,11 +145,13 @@ model VideoTranscript {
   audio_type         String?
   transcript         String[]
   non_speech_summary String?
+  content_tags       String[]                 @default([])
   createdAt          DateTime                 @default(now())
   updatedAt          DateTime                 @updatedAt
   transcript_zh_tsv  Unsupported("tsvector")?
   video              Video                    @relation(fields: [videoId], references: [aweme_id], onDelete: Cascade)
 
   @@index([videoId])
+  @@index([content_tags], map: "video_transcript_content_tags_idx", type: Gin)
   @@index([transcript_zh_tsv], map: "video_transcript_zh_idx", type: Gin)
 }

scripts/asr-cli.cjs

@@ -0,0 +1,135 @@
+#!/usr/bin/env node
+const fs = require("node:fs");
+const Module = require("node:module");
+const path = require("node:path");
+
+const rootDir = path.resolve(__dirname, "..");
+
+function loadEnvFile(filePath, override = false) {
+  if (!fs.existsSync(filePath)) return;
+
+  const lines = fs.readFileSync(filePath, "utf8").split(/\r?\n/);
+  for (const line of lines) {
+    const trimmed = line.trim();
+    if (!trimmed || trimmed.startsWith("#")) continue;
+
+    const normalized = trimmed.startsWith("export ")
+      ? trimmed.slice("export ".length).trim()
+      : trimmed;
+    const match = normalized.match(/^([A-Za-z_][A-Za-z0-9_]*)\s*=\s*(.*)$/);
+    if (!match) continue;
+
+    const key = match[1];
+    let value = match[2].trim();
+    if (
+      (value.startsWith('"') && value.endsWith('"')) ||
+      (value.startsWith("'") && value.endsWith("'"))
+    ) {
+      value = value.slice(1, -1);
+    }
+    value = value.replace(/\\n/g, "\n");
+
+    if (override || process.env[key] === undefined) {
+      process.env[key] = value;
+    }
+  }
+}
+
+function registerTypeScriptRuntime() {
+  const ts = require("typescript");
+  const originalResolveFilename = Module._resolveFilename;
+
+  Module._resolveFilename = function resolveFilename(
+    request,
+    parent,
+    isMain,
+    options,
+  ) {
+    if (request.startsWith("@/")) {
+      return originalResolveFilename.call(
+        this,
+        path.join(rootDir, request.slice(2)),
+        parent,
+        isMain,
+        options,
+      );
+    }
+    return originalResolveFilename.call(this, request, parent, isMain, options);
+  };
+
+  const compile = (module, filename) => {
+    const source = fs.readFileSync(filename, "utf8");
+    const output = ts.transpileModule(source, {
+      fileName: filename,
+      compilerOptions: {
+        module: ts.ModuleKind.CommonJS,
+        target: ts.ScriptTarget.ES2022,
+        esModuleInterop: true,
+        resolveJsonModule: true,
+        jsx: ts.JsxEmit.ReactJSX,
+        moduleResolution: ts.ModuleResolutionKind.Node10,
+      },
+    }).outputText;
+    module._compile(output, filename);
+  };
+
+  require.extensions[".ts"] = compile;
+  require.extensions[".tsx"] = compile;
+  require.extensions[".md"] = (module, filename) => {
+    module.exports = fs.readFileSync(filename, "utf8");
+  };
+}
+
+function getAwemeId() {
+  const args = process.argv.slice(2);
+  const fromFlag = args.find((arg) => arg.startsWith("--aweme-id="));
+  if (fromFlag) return fromFlag.slice("--aweme-id=".length).trim();
+
+  const flagIndex = args.findIndex(
+    (arg) => arg === "--aweme-id" || arg === "-a",
+  );
+  if (flagIndex >= 0) return (args[flagIndex + 1] || "").trim();
+
+  return (args.find((arg) => !arg.startsWith("-")) || "").trim();
+}
+
+function usage() {
+  console.log("用法: npm run asr -- <aweme_id>");
+  console.log("或:   node scripts/asr-cli.cjs --aweme-id <aweme_id>");
+}
+
+function logStep(message) {
+  const now = new Date().toISOString();
+  console.log(`[asr ${now}] ${message}`);
+}
+
+async function main() {
+  loadEnvFile(path.join(rootDir, ".env"));
+  loadEnvFile(path.join(rootDir, ".env.local"), true);
+  registerTypeScriptRuntime();
+
+  const awemeId = getAwemeId();
+  if (!awemeId) {
+    usage();
+    process.exitCode = 1;
+    return;
+  }
+
+  logStep(`开始执行 aweme_id=${awemeId}`);
+
+  const { transcriptAweme } = require(path.join(rootDir, "app/api/asr"));
+  const { prisma } = require(path.join(rootDir, "lib/prisma"));
+
+  try {
+    const result = await transcriptAweme(awemeId, { log: logStep });
+    logStep("最终结果:");
+    console.log(JSON.stringify(result, null, 2));
+  } finally {
+    await prisma.$disconnect();
+  }
+}
+
+main().catch((error) => {
+  console.error("[asr] 执行失败:", error);
+  process.exitCode = 1;
+});