Update app.py

app.py

@@ -39,7 +39,13 @@ def transcribe(audio):
         device=device,
     )
     
-    prediction = pipe(audio_np)["text"]
+    # Create an attention mask
+    attention_mask = np.ones(audio_np.shape, dtype=np.int64)
+    
+    # Use EncoderDecoderCache for past_key_values
+    past_key_values = EncoderDecoderCache.from_legacy_cache(None)
+    
+    prediction = pipe(audio_np, attention_mask=attention_mask, past_key_values=past_key_values)["text"]
     return prediction
     
 gradio_app = gr.Interface(