使用 qwen 进行语音识别，加入 tag

Co-authored-by: Copilot <copilot@github.com>
2026-04-29 20:10:08 +08:00
parent 040273586b
commit 206a561ee7
26 changed files with 3720 additions and 443 deletions
@@ -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 { NextResponse } from "next/server";
+import { prisma } from "@/lib/prisma";
+
+type SearchRow = {
+  id: string;
+  awemeId: string;
+  type: "video" | "image";
+  rank: number;
+  snippet: string;
+  contentTags: string[];
+};
+
+function sanitizeSnippetHtml(snippet: string) {
+  return snippet
+    .replaceAll("&", "&amp;")
+    .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<
-      {
-        id: string;
-        awemeId: string;
-        type: "video" | "image";
-        rank: number;
-        snippet: string;
-      }[]
-    >`
-      WITH tsq AS (
-        SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
-      )
-      -- 视频搜索结果
-      SELECT
-        COALESCE(vt.id, 'desc:' || v.aweme_id) AS id,
-        v.aweme_id AS "awemeId",
-        'video'::text AS type,
-        GREATEST(
-          COALESCE(ts_rank(vt.transcript_zh_tsv, tsq.query), 0),
-          COALESCE(ts_rank(to_tsvector('zhcfg', v.desc), tsq.query), 0)
-        ) AS rank,
-        CASE
-          WHEN vt.transcript_zh_tsv @@ tsq.query THEN
+    const rows = q
+      ? await prisma.$queryRaw<SearchRow[]>`
+          WITH tsq AS (
+            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
+          )
+          -- 视频搜索结果
+          SELECT
+            COALESCE(vt.id, 'desc:' || v.aweme_id) AS id,
+            v.aweme_id AS "awemeId",
+            'video'::text AS type,
+            GREATEST(
+              COALESCE(ts_rank(vt.transcript_zh_tsv, tsq.query), 0),
+              COALESCE(ts_rank(to_tsvector('zhcfg', v.desc), tsq.query), 0)
+            ) AS rank,
+            CASE
+              WHEN vt.transcript_zh_tsv @@ tsq.query THEN
+                ts_headline(
+                  'zhcfg',
+                  array_to_string(vt.transcript, ' '),
+                  tsq.query,
+                  'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
+                )
+              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'
-            )
-          ELSE
-            ts_headline(
-              'zhcfg',
+            ) AS snippet,
+            ARRAY[]::text[] AS "contentTags"
+          FROM "ImagePost" ip, tsq
+          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,
-              'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
-            )
-        END AS snippet
-      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)
-      
-      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 ${limit};
-    `;
+              '标签匹配'
+            ) AS snippet,
+            vt.content_tags AS "contentTags"
+          FROM "VideoTranscript" vt
+          JOIN "Video" v ON v.aweme_id = vt."videoId"
+          WHERE vt.content_tags @> ARRAY[${contentTag}]::text[]
+          ORDER BY v.created_at DESC
+          OFFSET ${offset}
+          LIMIT ${limit};
+        `;

    // 查询总数（参数化，避免注入）
-    const totalRows = await prisma.$queryRaw<
-      {
-        count: number;
-      }[]
-    >`
-      WITH tsq AS (
-        SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
-      )
-      SELECT (
-        (SELECT COUNT(*)::int FROM "Video" v
-         LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
-         WHERE (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query))
-        +
-        (SELECT COUNT(*)::int FROM "ImagePost" ip
-         WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query)
-      ) AS count
-      FROM tsq;
-    `;
+    const totalRows = q
+      ? await prisma.$queryRaw<{ count: number }[]>`
+          WITH tsq AS (
+            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
+          )
+          SELECT (
+            (SELECT COUNT(*)::int FROM "Video" v
+             LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
+             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)
+            +
+            (SELECT COUNT(*)::int FROM "ImagePost" ip
+             WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
+               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)