优化输出结果

asr_service.py

@@ -183,7 +183,7 @@ class ASRService:
             asr_begin = asr_seg["begin_time"]
             asr_end = asr_seg["end_time"]
 
-            best_speaker = "SPEAKER_00"
+            best_speaker = "speaker_0"
             best_overlap = 0.0
 
             for dia_seg in diarization_segments:

main.py

@@ -282,10 +282,10 @@ def process_batch_asr(video_paths: List[Path], diar_results: Dict, max_workers:
                     continue
 
             # 2. 加载说话人分离结果
-            from map_speaker import find_speaker, load_json
+            from map_speaker import load_json
             diar_result = load_json(diar_path)
 
-            # 3. 执行 ASR 识别
+            # 3. 执行 ASR 识别（不使用 ASR 自带的说话人）
             asr_sentences = asr_service.recognize(wav_path)
 
             if not asr_sentences:
@@ -297,15 +297,32 @@ def process_batch_asr(video_paths: List[Path], diar_results: Dict, max_workers:
                 })
                 continue
 
-            # 4. 合并说话人信息
+            # 4. 合并说话人信息（只使用 3D-Speaker 结果）
             print(f"  合并结果...")
             for sentence in asr_sentences:
-                new_speaker = find_speaker(
-                    sentence.begin_time,
-                    sentence.end_time,
-                    diar_result["segments"]
-                )
-                sentence.speaker = new_speaker
+                # 查找该时间段对应的说话人
+                matched_speaker = None
+                best_overlap = 0.0
+
+                for seg in diar_result["segments"]:
+                    seg_begin = seg['begin_time']
+                    seg_end = seg['end_time']
+
+                    # 计算重叠时间
+                    overlap_begin = max(sentence.begin_time, seg_begin)
+                    overlap_end = min(sentence.end_time, seg_end)
+
+                    if overlap_begin < overlap_end:
+                        overlap_duration = overlap_end - overlap_begin
+                        if overlap_duration > best_overlap:
+                            best_overlap = overlap_duration
+                            matched_speaker = seg['speaker']
+
+                # 如果有匹配，使用匹配的说话人；否则使用 speaker_0
+                if matched_speaker:
+                    sentence.speaker = matched_speaker
+                else:
+                    sentence.speaker = "speaker_0"
 
             # 5. 保存最终结果
             output_file = OUTPUT_DIR / f"{video_path.stem}_result.json"