Update tasks/audio.py

tasks/audio.py

@@ -60,47 +60,49 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     model_path = "quantized_teacher_m5_static.pth"
     model, device = load_model(model_path)
 
-    def preprocess_audio(example, target_length=32000):
-        """
-        Convert dataset into tensors:
-        - Convert to tensor
-        - Normalize waveform
-        - Pad/truncate to `target_length`
-        """
-        waveform = torch.tensor(example["audio"]["array"], dtype=torch.float32).unsqueeze(0)  # Add batch dim
+    # def preprocess_audio(example, target_length=32000):
+    #     """
+    #     Convert dataset into tensors:
+    #     - Convert to tensor
+    #     - Normalize waveform
+    #     - Pad/truncate to `target_length`
+    #     """
+    #     waveform = torch.tensor(example["audio"]["array"], dtype=torch.float32).unsqueeze(0)  # Add batch dim
     
-        # Normalize waveform
-        waveform = (waveform - waveform.mean()) / (waveform.std() + 1e-6)
+    #     # Normalize waveform
+    #     waveform = (waveform - waveform.mean()) / (waveform.std() + 1e-6)
     
-        # Pad or truncate to fixed length
-        if waveform.shape[1] < target_length:
-            pad = torch.zeros(1, target_length - waveform.shape[1])
-            waveform = torch.cat((waveform, pad), dim=1)  # Pad
-        else:
-            waveform = waveform[:, :target_length]  # Truncate
+    #     # Pad or truncate to fixed length
+    #     if waveform.shape[1] < target_length:
+    #         pad = torch.zeros(1, target_length - waveform.shape[1])
+    #         waveform = torch.cat((waveform, pad), dim=1)  # Pad
+    #     else:
+    #         waveform = waveform[:, :target_length]  # Truncate
     
-        label = torch.tensor(example["label"], dtype=torch.long)  # Ensure int64
-    return {"waveform": waveform, "label": label}
+    #     label = torch.tensor(example["label"], dtype=torch.long)  # Ensure int64
+    # return {"waveform": waveform, "label": label}
 
 
 
-    train_test = train_test.map(preprocess_audio, batched=True)
-    test_dataset = train_test.map(preprocess_audio)
+    # train_test = train_test.map(preprocess_audio, batched=True)
+    # test_dataset = train_test.map(preprocess_audio)
 
-    train_loader = DataLoader(train_test, batch_size=32, shuffle=True)
+    # train_loader = DataLoader(train_test, batch_size=32, shuffle=True)
 
     
     true_labels = train_dataset["label"]
     predictions = []
+   
+    predictions = [random.randint(0, 1) for _ in range(len(true_labels))]
 
-    with torch.no_grad():
-        for waveforms, labels in train_loader:
-            waveforms, labels = waveforms.to(device), labels.to(device)
+    # with torch.no_grad():
+    #     for waveforms, labels in train_loader:
+    #         waveforms, labels = waveforms.to(device), labels.to(device)
             
-            outputs = model(waveforms)
-            predicted_label = torch.argmax(F.softmax(outputs, dim=1), dim=1)
-            true_labels.extend(labels.cpu().numpy())
-            predicted_labels.extend(predicted_label.cpu().numpy())
+    #         outputs = model(waveforms)
+    #         predicted_label = torch.argmax(F.softmax(outputs, dim=1), dim=1)
+    #         true_labels.extend(labels.cpu().numpy())
+    #         predicted_labels.extend(predicted_label.cpu().numpy())
     
     #--------------------------------------------------------------------------------------------
     # YOUR MODEL INFERENCE STOPS HERE