app.py

import gradio as gr
from transformers import pipeline
import soundfile as sf
import os

# --- Model Loading ---
# We switched to 'superb/wav2vec2-base-superb-er' as it's a well-established and public model for emotion recognition.
try:
    classifier = pipeline("audio-classification", model="superb/wav2vec2-base-superb-er")
except Exception as e:
    # If there's an error during model loading, we can display it in the Gradio interface
    # This helps in debugging issues directly on the Hugging Face Space.
    def error_fn(audio_file):
        return {"error": f"Failed to load the model. Please check the logs. Error: {str(e)}"}
    classifier = None

# --- Prediction Function ---
def predict_emotion(audio_file):
    """
    Predicts emotions from an audio file.

    Args:
        audio_file (str or tuple): Path to the audio file (from upload) or a tuple
                                   (samplerate, audio_array) from microphone input.
    Returns:
        dict: A dictionary of emotion labels and their probabilities.
    """
    # Handle case where the model failed to load
    if classifier is None:
        return {"error": "The AI model could not be loaded. The application cannot start."}
    
    if audio_file is None:
        return {"error": "No audio input provided. Please upload a file or record."}

    # Gradio's Audio component can return a path to a temp file for file uploads,
    # or a tuple (samplerate, numpy_array) for microphone input.
    if isinstance(audio_file, str):
        # Handle file path (e.g., from file upload)
        audio_path = audio_file
    elif isinstance(audio_file, tuple):
        # Handle microphone input (samplerate, numpy_array)
        sample_rate, audio_array = audio_file
        # Save the numpy array to a temporary WAV file as the pipeline expects a file path or direct bytes
        temp_audio_path = "temp_audio_from_mic.wav"
        sf.write(temp_audio_path, audio_array, sample_rate)
        audio_path = temp_audio_path
    else:
        return {"error": f"Invalid audio input format: {type(audio_file)}"}

    try:
        # Perform inference
        results = classifier(audio_path, top_k=5) # top_k ensures we get all relevant emotion scores

        # Process results into a dictionary for better display
        emotion_scores = {item['label']: round(item['score'], 3) for item in results}

        return emotion_scores
    except Exception as e:
        return {"error": f"An error occurred during prediction: {str(e)}"}
    finally:
        # Clean up temporary file if created
        if 'temp_audio_path' in locals() and os.path.exists(temp_audio_path):
            os.remove(temp_audio_path)


# --- Gradio Interface ---
# Define the Gradio interface
iface = gr.Interface(
    fn=predict_emotion,
    # THIS IS THE CORRECTED LINE:
    inputs=gr.Audio(sources=["microphone", "upload"], type="filepath", label="Upload Audio or Record with Microphone"),
    outputs=gr.Label(num_top_classes=5, label="Emotion Probabilities"), # This model has 4 emotions + 'no-emotion'
    title="AI Audio Emotion Detector",
    description="Upload an audio file or record your voice to detect emotions. This model is trained to recognize 'anger', 'happiness', 'neutral', 'sadness', and 'no-emotion'.",
    examples=[
        # You can add example audio files to your Hugging Face Space and reference them here.
        # For now, we'll leave this empty.
    ]
)

# Launch the Gradio app
if __name__ == "__main__":
    iface.launch()