Update app.py

app.py

@@ -14,9 +14,13 @@ def transcribe(audio):
     # If audio is a tuple, extract the NumPy array and the sampling rate
     if isinstance(audio, tuple):
         audio_input, sr = audio
+        # Save the audio back to a file for playback
+        audio_file = "temp.wav"
+        librosa.output.write_wav(audio_file, audio_input, sr)
     else:
         # Otherwise, load the file from the path
-        audio_input, sr = librosa.load(audio, sr=16000)
+        audio_file = audio
+        audio_input, sr = librosa.load(audio_file, sr=16000)
     
     # Ensure the sample rate is what the processor expects
     if sr != 16000:
@@ -27,12 +31,13 @@ def transcribe(audio):
     predicted_ids = model.generate(input_features)
     transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
     text = transcription[0]  # Decode returns a list
-    return text
+    
+    return audio_file, text
 
 iface = gr.Interface(
     fn=transcribe,
-    inputs=gr.Audio(),  # No need for 'source' or 'type'
-    outputs="text",
+    inputs=gr.Audio(),  # Audio input
+    outputs=[gr.Audio(), "text"],  # Return the audio file and transcription text
     title="Whisper Medium Shqip",
     description="Realtime demo for Sq speech recognition using a fine-tuned Whisper medium model.",
 )