app.py

import gradio as gr
import torch
from transformers import pipeline
import spaces  # Ensure spaces library is imported if using GPU decorator

@spaces.GPU(duration=3)  # Decorator to allocate GPU for the app
def main():
    # Force GPU if available, fallback to CPU
    device = 0 if torch.cuda.is_available() else -1

    try:
        # Load models with explicit device
        transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
        summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
    except Exception as e:
        print(f"Error loading models: {e}")
        raise

    # Function to process audio
    def process_audio(audio_file):
        try:
            # Transcribe the audio
            transcription = transcriber(audio_file)["text"]
            # Summarize the transcription
            summary = summarizer(transcription, max_length=50, min_length=10, do_sample=False)[0]["summary_text"]
            return transcription, summary
        except Exception as e:
            return f"Error processing audio: {e}", ""

    # Gradio Interface with Horizontal Layout
    with gr.Blocks() as interface:
        with gr.Row():
            with gr.Column():
                audio_input = gr.Audio(type="filepath", label="Upload Audio File")
                process_button = gr.Button("Process Audio")
            with gr.Column():
                transcription_output = gr.Textbox(label="Full Transcription", lines=10)
                summary_output = gr.Textbox(label="Summary", lines=5)

        process_button.click(
            process_audio,
            inputs=[audio_input],
            outputs=[transcription_output, summary_output]
        )

    # Launch the interface with optional public sharing
    interface.launch(share=True)

# Run the main function
if __name__ == "__main__":
    main()