feat: 将视频转写重构为纯文字异步语音识别

2026-10-03 21:58:06 +08:00
parent 040273586b
commit 6f0e38054b
34 changed files with 2283 additions and 1771 deletions
@@ -0,0 +1,47 @@
+BEGIN;
+ALTER TABLE "VideoTranscript"
+  DROP COLUMN IF EXISTS speech_detected,
+  DROP COLUMN IF EXISTS language,
+  DROP COLUMN IF EXISTS audio_type,
+  DROP COLUMN IF EXISTS non_speech_summary,
+  DROP COLUMN IF EXISTS content_tags;
+
+CREATE TABLE IF NOT EXISTS "TranscriptionTask" (
+  "videoId" TEXT PRIMARY KEY REFERENCES "Video"(aweme_id) ON DELETE CASCADE ON UPDATE CASCADE,
+  status TEXT NOT NULL DEFAULT 'QUEUED' CHECK (status IN ('QUEUED', 'PROCESSING', 'SUCCEEDED', 'FAILED')),
+  "providerTaskId" TEXT,
+  "submittingAt" TIMESTAMP(3),
+  "submittedAt" TIMESTAMP(3),
+  "nextPollAt" TIMESTAMP(3) NOT NULL DEFAULT (CURRENT_TIMESTAMP AT TIME ZONE 'UTC'),
+  "leaseUntil" TIMESTAMP(3),
+  "leaseToken" TEXT,
+  attempts INTEGER NOT NULL DEFAULT 0,
+  "lastError" TEXT,
+  "createdAt" TIMESTAMP(3) NOT NULL DEFAULT (CURRENT_TIMESTAMP AT TIME ZONE 'UTC'),
+  "updatedAt" TIMESTAMP(3) NOT NULL DEFAULT (CURRENT_TIMESTAMP AT TIME ZONE 'UTC')
+);
+CREATE INDEX IF NOT EXISTS "TranscriptionTask_status_nextPollAt_idx" ON "TranscriptionTask"(status, "nextPollAt");
+
+-- Preserve an existing Chinese parser configuration. Fresh databases use simple
+-- until the operator installs a Chinese parser and configures public.zhcfg.
+DO $$ BEGIN
+  IF NOT EXISTS (SELECT 1 FROM pg_ts_config c JOIN pg_namespace n ON n.oid = c.cfgnamespace
+                 WHERE n.nspname = 'public' AND c.cfgname = 'zhcfg') THEN
+    CREATE TEXT SEARCH CONFIGURATION public.zhcfg (COPY = pg_catalog.simple);
+  END IF;
+END $$;
+ALTER TABLE "VideoTranscript" ADD COLUMN IF NOT EXISTS transcript_zh_tsv tsvector;
+CREATE OR REPLACE FUNCTION public.refresh_video_transcript_search() RETURNS trigger AS $$
+BEGIN
+  NEW.transcript_zh_tsv := to_tsvector('public.zhcfg'::regconfig, COALESCE(array_to_string(NEW.transcript, ' '), ''));
+  RETURN NEW;
+END;
+$$ LANGUAGE plpgsql;
+-- Replace the legacy live trigger as well as our own trigger on repeat deployment.
+DROP TRIGGER IF EXISTS trg_vt_transcript_zh_tsv ON "VideoTranscript";
+DROP TRIGGER IF EXISTS video_transcript_search_update ON "VideoTranscript";
+CREATE TRIGGER video_transcript_search_update BEFORE INSERT OR UPDATE OF transcript
+  ON "VideoTranscript" FOR EACH ROW EXECUTE FUNCTION public.refresh_video_transcript_search();
+UPDATE "VideoTranscript" SET transcript_zh_tsv = to_tsvector('public.zhcfg'::regconfig, COALESCE(array_to_string(transcript, ' '), ''));
+CREATE INDEX IF NOT EXISTS video_transcript_zh_idx ON "VideoTranscript" USING GIN (transcript_zh_tsv);
+COMMIT;