Update app.py

app.py

@@ -81,21 +81,47 @@ def format_transcript_with_speakers(transcript, diarization):
     return "".join(formatted_transcript)
 
 def transcribe_audio(audio_file):
-    # Perform diarization on the entire audio file
-    diarization = pipeline(audio_file)
-
-    # Load the audio
-    audio_input, sr = librosa.load(audio_file, sr=16000)
-
-    # Transcribe the entire audio (or use chunking with time tracking if necessary)
-    input_features = processor(audio_input, sampling_rate=16000, return_tensors="pt").input_features.to(device)
-    predicted_ids = model.generate(input_features)
-    full_transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
+    try:
+        print("Loading audio file...")
+        audio_input, sr = librosa.load(audio_file, sr=16000)
+        audio_input = audio_input.astype(np.float32)
+        print(f"Audio duration: {len(audio_input) / sr:.2f} seconds")
+
+        # Apply speaker diarization
+        if pipeline:
+            print("Applying speaker diarization...")
+            diarization = pipeline(audio_file)
+            print("Speaker diarization complete.")
+        else:
+            diarization = None
 
-    # Apply diarization to the full transcription
-    formatted_transcription = format_transcript_with_speakers(full_transcription, diarization)
+        chunk_length = 30 * sr
+        overlap = 5 * sr
+        transcriptions = []
+        
+        print("Starting transcription...")
+        for i in range(0, len(audio_input), chunk_length - overlap):
+            chunk = audio_input[i:i+chunk_length]
+            input_features = processor(chunk, sampling_rate=16000, return_tensors="pt").input_features.to(device)
+            predicted_ids = model.generate(input_features)
+            transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
+            transcriptions.extend(transcription)
+            print(f"Processed {i / sr:.2f} to {(i + chunk_length) / sr:.2f} seconds")
+
+        full_transcription = " ".join(transcriptions)
+        print(f"Transcription complete. Full transcription length: {len(full_transcription)} characters")
+
+        if diarization:
+            print("Applying formatting with speaker diarization...")
+            formatted_transcription = format_transcript_with_speakers(full_transcription, diarization)
+        else:
+            print("Applying formatting without speaker diarization...")
+            formatted_transcription = format_transcript_with_breaks(full_transcription)
 
-    return formatted_transcription
+        return formatted_transcription
+    except Exception as e:
+        print(f"Error in transcribe_audio: {str(e)}")
+        raise
 
 def format_transcript_with_breaks(transcript):
     sentences = re.split('(?<=[.!?]) +', transcript)