Update app.py

app.py

@@ -71,7 +71,6 @@ model = CNN1DLSTMAudioClassifier(num_class)
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model.load_state_dict(torch.load("speech-emotion-recognition-best-model.bin", weights_only=False))
-model = model.to(device)
 model.eval()
 
 def preprocess_single_audio(file_path, sample_rate=16000, n_mels=128, n_fft=2048, hop_length=512):
@@ -143,6 +142,7 @@ def predict(wave):
     le = LabelEncoder()
     le.classes_ = np.array(['Angry', 'Disgusting', 'Fear', 'Happy', 'Neutral', 'Sad'])
     wave = wave.unsqueeze(0)
+    wave = wave.to(device)
     with torch.no_grad():
         prediction = model(wave)
     predicted_emotion, confidence = decode_emotion_prediction(prediction, le)