Update app.py

app.py

@@ -13,7 +13,11 @@ def custom_feature_extraction(audio, sr=16000, n_mels=128, n_fft=2048, hop_lengt
     S_DB = librosa.power_to_db(S, ref=np.max)
     pitches, _ = librosa.piptrack(y=audio, sr=sr, n_fft=n_fft, hop_length=hop_length)
     spectral_centroids = librosa.feature.spectral_centroid(y=audio, sr=sr, n_fft=n_fft, hop_length=hop_length)
-    features = np.concatenate([S_DB, pitches.max(axis=1, keepdims=True), spectral_centroids], axis=0)
+    # Correct the dimensionality issue
+    pitches_max = np.max(pitches, axis=0, keepdims=True)
+    spectral_centroids = spectral_centroids.reshape(1, -1)
+    # Ensure the concatenation axis has matching dimensions
+    features = np.concatenate([S_DB, pitches_max, spectral_centroids], axis=0)
     features_tensor = torch.tensor(features).float()
     if features_tensor.shape[1] > target_length:
         features_tensor = features_tensor[:, :target_length]
@@ -51,3 +55,4 @@ iface = gr.Interface(
 )
 
 iface.launch()
+