Update app.py

app.py

@@ -17,9 +17,10 @@ except Exception as e:
 else:
     model_load_error = None
 
-# --- FastAPI App for a dedicated, robust API ---
+# --- FastAPI App ---
 app = FastAPI()
 
+# This is our dedicated, robust API endpoint
 @app.post("/api/predict/")
 async def predict_emotion_api(request: Request):
     if classifier is None:
@@ -39,38 +40,52 @@ async def predict_emotion_api(request: Request):
             header, encoded = base64_with_prefix.split(",", 1)
             audio_data = base64.b64decode(encoded)
         except (ValueError, TypeError):
-             return JSONResponse(content={"error": "Invalid base64 data format."}, status_code=400)
+             return JSONResponse(content={"error": "Invalid base64 data format. Please send the full data URI."}, status_code=400)
 
-        # Write to a temporary file for the pipeline
+        # Write to a temporary file for the pipeline to process
         with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_file:
             temp_file.write(audio_data)
             temp_audio_path = temp_file.name
 
-        results = classifier(temp_audio_path, top_k=5)
+        results = classifier(temp_audio_path)
         os.unlink(temp_audio_path) # Clean up the temp file
 
-        # Return a successful response
+        # The transformers pipeline returns a list of dicts
+        # Example: [{'score': 0.99, 'label': 'happy'}, {'score': 0.01, 'label': 'sad'}]
+        # We will return this directly
         return JSONResponse(content={"data": results})
 
     except Exception as e:
+        # Clean up the temp file if it exists even after an error
+        if 'temp_audio_path' in locals() and os.path.exists(temp_audio_path):
+            os.unlink(temp_audio_path)
         return JSONResponse(content={"error": f"Internal server error during prediction: {str(e)}"}, status_code=500)
 
-# --- Gradio UI function (optional, for the direct Space page) ---
-def gradio_predict_wrapper(audio_file):
-    # This is just for the UI on the Hugging Face page itself
-    if audio_file is None: return {"error": "Please provide an audio file."}
-    results = classifier(audio_file, top_k=5)
-    return {item['label']: round(item['score'], 3) for item in results}
+# --- Gradio UI (for demonstration on the Space's page) ---
+def gradio_predict_wrapper(audio_file_path):
+    if classifier is None: return {"error": f"Model is not loaded: {model_load_error}"}
+    if audio_file_path is None: return {"error": "Please provide an audio file."}
+    
+    try:
+        results = classifier(audio_file_path, top_k=5)
+        # Format for Gradio's Label component
+        return {item['label']: item['score'] for item in results}
+    except Exception as e:
+        return {"error": str(e)}
 
 gradio_interface = gr.Interface(
     fn=gradio_predict_wrapper,
     inputs=gr.Audio(sources=["microphone", "upload"], type="filepath", label="Upload Audio or Record"),
     outputs=gr.Label(num_top_classes=5, label="Emotion Predictions"),
     title="Audio Emotion Detector",
-    description="This UI is for direct demonstration. The primary API is at /api/predict/",
+    description="This UI is for direct demonstration. The primary API for websites is at /api/predict/",
     allow_flagging="never"
 )
 
-# --- Mount the Gradio UI onto the FastAPI app ---
-# The API at /api/predict/ will work even if the UI is at a different path.
-app = gr.mount_gradio_app(app, gradio_interface, path="/ui")
+# Mount the Gradio UI onto a subpath of our FastAPI app
+app = gr.mount_gradio_app(app, gradio_interface, path="/gradio")
+
+# The Uvicorn server launch command (used by Hugging Face Spaces)
+# This is the ONLY launch command needed.
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=7860)