首次提交

tools/gen_test_srt.py

@@ -0,0 +1,47 @@
+# -*- coding: utf-8 -*-
+"""用真实OCR数据生成一份【已过滤水印】的测试SRT。跳过读图(monkeypatch)。
+用法: python temp/gen_test_srt.py <composite目录> <输出srt>
+"""
+import sys, os, json
+sys.path.insert(0, os.path.abspath("."))
+import tool.getsrt as G
+from tool.log.log_manager import UnifiedLogger
+
+G.find_best_ocr_match = lambda *a, **k: (None, 0)   # 跳过首尾音频匹配的读图
+
+D = sys.argv[1] if len(sys.argv) > 1 else "test_srt/BV1BamdYLExJ"
+OUT = sys.argv[2] if len(sys.argv) > 2 else "test_srt/BV1BamdYLExJ.filtered.srt"
+META = os.path.join(D, "metadata.json")
+logger = UnifiedLogger(); logger.reset(); logger.init_logging("rerun_log.json"); logger.start_program("rerun")
+
+pairs = []
+lines = open(os.path.join(D, "ocr_result.jsonl"), encoding="utf-8").readlines()
+i = 0
+while i < len(lines):
+    if i + 1 < len(lines) and lines[i].startswith("pic_path: "):
+        fn = os.path.basename(lines[i].strip()[len("pic_path: "):])
+        try:
+            pairs.append([os.path.join(D, fn), json.loads(lines[i + 1])]); i += 2; continue
+        except json.JSONDecodeError:
+            pass
+    i += 1
+
+# 坐标水印过滤(就是接进 getsrt 的那个)
+G._filter_watermark_by_position(pairs, logger)
+
+best = G.analyze_subtitle_blocks(pairs, META, logger)
+h = G.analyze_h(pairs, META)
+results = []
+for idx in range(len(pairs)):
+    try:
+        text, conf, flag, sim = G.extract_highest_subtitle(idx, pairs, best, META, h, logger)
+    except Exception:
+        text = ""
+    if text and not G.check_subtitle_quality_enhanced(text) and conf <= 90:
+        text = ""
+    results.append((idx, text or ""))
+# 静态文字过滤(频率, 兜底)
+results = G._filter_static_text(results, logger)
+G.generate_srt_from_results(results, OUT, interval_seconds=0.33)
+n = sum(1 for l in open(OUT, encoding="utf-8") if "-->" in l)
+print("生成: %s  共 %d 条字幕" % (OUT, n))