# audio_classification_app.py
import gradio as gr
from transformers import pipeline

# Load pipeline (pretrained model)
classifier = pipeline("audio-classification", model="superb/wav2vec2-base-superb-ks")

def classify_audio(audio):
    if audio is None:
        return {"error": "No audio provided"}
    
    # Gradio audio returns (sample_rate, data)
    sr, data = audio
    
    # Convert to Hugging Face expected format
    inputs = {"array": data, "sampling_rate": sr}
    
    # Run classification
    results = classifier(inputs)
    return results

with gr.Blocks() as demo:
    gr.Markdown("## 🎵 Audio Classification (Keyword Spotting)")
    with gr.Row():
        audio_input = gr.Audio(sources=["microphone", "upload"], type="numpy")
        output = gr.JSON()
    audio_input.change(classify_audio, audio_input, output)

if __name__ == "__main__":
    demo.launch()