Enhance transcription segment coverage calculation: Updated the overlap check to consider total coverage from all transcription segments, ensuring segments are re-transcribed if less than 85% of their duration is covered. This improves accuracy in identifying segments needing attention.

app.py

@@ -726,7 +726,16 @@ class WhisperTranscriber:
             d_end = float(dseg["end"])
             has_overlap = False
             for seg in transcription_results:
-                if interval_overlap(d_start, d_end, float(seg["start"]), float(seg["end"])) > abs(d_start - d_end) * 0.6:
+                # Instead of just checking if any single segment covers 60%,
+                # let's calculate total coverage from ALL transcription segments
+                total_coverage = 0.0
+                for s in transcription_results:
+                    overlap = interval_overlap(d_start, d_end, float(s["start"]), float(s["end"]))
+                    total_coverage += overlap
+
+                coverage_ratio = total_coverage / (d_end - d_start)
+                if coverage_ratio < 0.85:  # Less than 80% covered
+                    # This diarization segment needs re-transcription
                     has_overlap = True
                     break
             if not has_overlap: