Update app.py

app.py

@@ -12,11 +12,34 @@ feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name_or_path)
 sampling_rate = feature_extractor.sampling_rate
 model = Wav2Vec2ForSpeechClassification.from_pretrained(model_name_or_path)
 
-def speech_file_to_array_fn(path, sampling_rate):
-    speech_array, _sampling_rate = torchaudio.load(path)
-    resampler = torchaudio.transforms.Resample(_sampling_rate, sampling_rate)
-    speech = resampler(speech_array).squeeze().numpy()
-    return speech
+# def speech_file_to_array_fn(path, sampling_rate):
+#     speech_array, _sampling_rate = torchaudio.load(path)
+#     resampler = torchaudio.transforms.Resample(_sampling_rate, sampling_rate)
+#     speech = resampler(speech_array).squeeze().numpy()
+#     return speech
+
+def speech_file_to_array_fn(audio_path):
+    if audio_path is None:
+        return None  # Handle cases where no file is provided
+    
+    try:
+        # Check if the input is a file path (upload) or direct audio data (recording)
+        if isinstance(audio_path, str):  
+            speech_array, _sampling_rate = torchaudio.load(audio_path)
+        else:  
+            # If it's recorded audio, Gradio provides it as a NumPy array
+            speech_array = torch.tensor(audio_path)  
+            _sampling_rate = sampling_rate  # Use default sampling rate
+        
+        # Resample to match model requirements
+        resampler = torchaudio.transforms.Resample(orig_freq=_sampling_rate, new_freq=sampling_rate)
+        speech = resampler(speech_array).squeeze().numpy()
+        return speech
+    
+    except Exception as e:
+        print(f"Error processing audio: {e}")
+        return None
+
 
 def predict(audio_path):
     speech = speech_file_to_array_fn(audio_path, sampling_rate)