Added badges for repo and presentation

app.py

@@ -122,31 +122,22 @@ def preprocess_drive_audio(file_path, target_sample_rate=22050, n_mels=128, nois
     return final_tensor
 
 
-def run_local_inference(model_instance, file_path, inference_device, noise_std=0.0):
+def run_local_inference(model, file_path, device, noise_std=0.0):
     """
-    Runs a forward pass and returns both the top predicted class 
-    and a dictionary of all class probabilities.
+    Runs a forward pass
     """
+
     emotion_classes = ["Neutral", "Calm", "Happy", "Sad", "Angry", "Fearful", "Disgust", "Surprised"]
 
     input_tensor = preprocess_drive_audio(file_path, noise_std=noise_std)
-    input_tensor = input_tensor.to(inference_device)
+    input_tensor = input_tensor.to(device)
 
-    model_instance.eval()
+    model.eval()
     with torch.no_grad():
-        logits = model_instance(input_tensor)
-        
-        # Get the top prediction index
+        logits = model(input_tensor)
         predicted_index = torch.argmax(logits, dim=1).item()
-        top_class = emotion_classes[predicted_index]
-        
-        # Apply softmax to convert logits to probabilities between 0 and 1
-        probabilities = torch.softmax(logits, dim=1).squeeze().tolist()
 
-    # Map the probabilities to their respective class names for Gradio
-    prob_dict = {emotion_classes[i]: probabilities[i] for i in range(len(emotion_classes))}
-    
-    return top_class, prob_dict
+    return emotion_classes[predicted_index]
 
 
 # Load SOTA SupConFN
@@ -164,13 +155,27 @@ model = model.to(device)
 @spaces.GPU
 def predict_emotion(audio_path, noise_std):
     if audio_path is None:
-        return "No file uploaded", {"Error": 1.0}
+        return "Upload an audio file."
     
     try:
-        top_pred, prob_dict = run_local_inference(model, audio_path, device, noise_std=noise_std)
-        return top_pred, prob_dict
+        pred = run_local_inference(model, audio_path, device, noise_std=noise_std)
+        return pred
     except Exception as e:
-        return "Error", {f"Error: {str(e)}": 1.0}
+        return f"Error processing audio: {str(e)}"
+
+
+description_html = """
+Upload a short audio clip and test the architecture's resilience. The slider injects mathematically controlled Gaussian noise into the normalized Mel-spectrogram before classification.
+
+<div style="display: flex; gap: 10px; margin-top: 15px;">
+    <a href="https://github.com/gpetrousov/dl_assignment_demokritos" target="_blank">
+        <img src="https://img.shields.io/badge/GitHub-View_Repository-181717?style=for-the-badge&logo=github" alt="GitHub Repository" />
+    </a>
+    <a href="https://docs.google.com/presentation/d/1ShSmXY2OqCEbQQjYiLYCFc8Psu2bioECoFHhliiDEiQ/edit?usp=drive_link" target="_blank">
+        <img src="https://img.shields.io/badge/Presentation-View_Slides-F4B400?style=for-the-badge&logo=googleslides&logoColor=white" alt="Presentation Slides" />
+    </a>
+</div>
+"""
 
 demo = gr.Interface(
     fn=predict_emotion,
@@ -183,9 +188,9 @@ demo = gr.Interface(
         gr.Label(num_top_classes=8, label="Probability Distribution")
     ],
     title="Robust Speech Emotion Recognition (SupCon)",
-    description="Upload a short audio clip and test the architecture's resilience. The slider injects mathematically controlled Gaussian noise into the normalized Mel-spectrogram before classification."
+    description=description_html,
+    allow_flagging="never"
 )
 
-
 if __name__ == "__main__":
-    demo.launch(ssr_mode=False)
+    demo.launch(ssr=False)