Update app.py

app.py

@@ -1099,8 +1099,14 @@ async def transcribe(audio: UploadFile = File(...), similarity_threshold: float
                 # Update expanded_start for accurate timing logs
                 expanded_start += boundary_offset
                 log(f"   Updated expanded start: {expanded_start:.3f}s")
+                
+                # ALSO apply the boundary offset to WhisperX timing
+                original_start_time = start_time
+                start_time = max(0, start_time + boundary_offset)
+                end_time = max(start_time + 0.01, end_time)  # Ensure minimum 10ms duration
+                log(f"   Updated WhisperX timing: {original_start_time:.3f}s → {start_time:.3f}s (shifted +{boundary_offset:.3f}s)")
             
-            # Also extract WhisperX original timing for comparison
+            # Also extract WhisperX original timing for comparison (now using updated start_time)
             whisperx_audio_segment = extract_audio_segment(waveform, sample_rate, start_time, end_time, word_clean, verbose=False)
             
             # Detect phoneme from expanded audio segment