Update app.py

app.py

@@ -44,13 +44,25 @@ summarization_tokenizer = AutoTokenizer.from_pretrained("t5-base")
 def transcribe_audio(audio_file):
     if audio_file.endswith(".m4a"):
         audio_file = convert_to_wav(audio_file)
-
+    
     start_time = time.time()
-    output = whisper_pipeline(audio_file)
-    text = output["text"]
-    output_time = time.time() - start_time
 
+    # Prepare input and attention mask
+    inputs = whisper_pipeline.tokenizer(audio_file, return_tensors="pt", padding=True)
+    inputs = {k: v.to(device) for k, v in inputs.items()}
+
+    # Generate the transcription with attention_mask
+    output = whisper_pipeline.model.generate(
+        inputs['input_ids'], 
+        attention_mask=inputs['attention_mask']
+    )
+
+    # Decode the output
+    text = whisper_pipeline.tokenizer.decode(output[0], skip_special_tokens=True)
+    
+    output_time = time.time() - start_time
     result = f"Time taken: {output_time:.2f} seconds\nNumber of words: {len(text.split())}"
+
     return text, result
 
 # Clean and preprocess text for summarization