Update app.py

app.py

@@ -69,33 +69,41 @@ def download_audio_from_url(url):
 
 def transcribe_audio(audio_file):
     try:
-        print("Loading audio file...")
+        logger.info("Loading audio file...")
         audio = AudioSegment.from_file(audio_file)
         audio = audio.set_channels(1).set_frame_rate(16000)
         audio_array = torch.tensor(audio.get_array_of_samples()).float()
         
-        print(f"Audio duration: {len(audio) / 1000:.2f} seconds")
-        print("Starting transcription...")
+        logger.info(f"Audio duration: {len(audio) / 1000:.2f} seconds")
+        logger.info("Starting transcription...")
         input_features = whisper_processor(audio_array, sampling_rate=16000, return_tensors="pt").input_features.to(device)
         
         # Create attention mask
         attention_mask = torch.ones_like(input_features)
         
-        # Generate with specific parameters
-        predicted_ids = whisper_model.generate(
-            input_features,
-            attention_mask=attention_mask,
-            language='en',
-            task='translate'
-        )
-        transcription = whisper_processor.batch_decode(predicted_ids, skip_special_tokens=True)
+        max_retries = 3
+        for attempt in range(max_retries):
+            # Generate with specific parameters
+            predicted_ids = whisper_model.generate(
+                input_features,
+                attention_mask=attention_mask,
+                language='en',
+                task='translate',
+                temperature=0.7,  # Adjust temperature for potentially better results
+                num_beams=5,  # Increase beam search for potentially better results
+                max_length=448,  # Increase max length to allow for longer transcriptions
+            )
+            transcription = whisper_processor.batch_decode(predicted_ids, skip_special_tokens=True)
+            
+            logger.info(f"Transcription attempt {attempt + 1} complete. Length: {len(transcription[0])} characters")
+            if len(transcription[0]) >= 10:
+                return transcription[0]
+            else:
+                logger.warning(f"Transcription too short on attempt {attempt + 1}: {transcription[0]}")
         
-        print(f"Transcription complete. Length: {len(transcription[0])} characters")
-        if len(transcription[0]) < 10:
-            raise ValueError(f"Transcription too short: {transcription[0]}")
-        return transcription[0]
+        raise ValueError(f"Failed to generate a valid transcription after {max_retries} attempts")
     except Exception as e:
-        print(f"Error in transcribe_audio: {str(e)}")
+        logger.error(f"Error in transcribe_audio: {str(e)}")
         raise
 
 def separate_speakers(transcription):