try use diarization as clip_timestamp

app.py

@@ -916,9 +916,20 @@ class WhisperTranscriber:
                 wav_path, num_speakers, base_offset_s=base_offset_s
             )
 
+            # Convert diarization_segments to clip_timestamps format
+            # Format: "start,end,start,end,..." with timestamps relative to the file (subtract base_offset_s)
+            clip_timestamps_list = []
+            for seg in diarization_segments:
+                # Convert global timestamps back to local file timestamps
+                local_start = max(0.0, float(seg["start"]) - base_offset_s)
+                local_end = max(local_start, float(seg["end"]) - base_offset_s)
+                clip_timestamps_list.extend([str(local_start), str(local_end)])
+
+            clip_timestamps = ",".join(clip_timestamps_list) if clip_timestamps_list else None
+
             # Step 2: Transcribe full audio once
             transcription_results, detected_language = self.transcribe_full_audio(
-                wav_path, language, translate, prompt, batch_size, base_offset_s=base_offset_s, clip_timestamps=diarization_segments
+                wav_path, language, translate, prompt, batch_size, base_offset_s=base_offset_s, clip_timestamps=clip_timestamps
             )