app.py

import gradio as gr
import torch
from transformers import pipeline
import librosa
import soundfile as sf
import spaces
import os

def split_audio(audio_data, sr, chunk_duration=30):
    """Split audio into chunks of chunk_duration seconds."""
    chunks = []
    for start in range(0, len(audio_data), int(chunk_duration * sr)):
        end = start + int(chunk_duration * sr)
        chunks.append(audio_data[start:end])
    return chunks

def transcribe_long_audio(audio_input, transcriber, chunk_duration=30):
    """Transcribe long audio by splitting into smaller chunks."""
    try:
        if isinstance(audio_input, tuple):  # Recorded audio
            audio_data, sr = audio_input
        elif isinstance(audio_input, str):  # Uploaded file path
            audio_data, sr = librosa.load(audio_input, sr=None)
        else:
            raise ValueError("Unsupported audio input format.")

        chunks = split_audio(audio_data, sr, chunk_duration)
        transcriptions = []
        for i, chunk in enumerate(chunks):
            temp_path = f"temp_chunk_{i}.wav"
            sf.write(temp_path, chunk, sr)  # Save chunk as WAV
            transcription = transcriber(temp_path)["text"]
            transcriptions.append(transcription)
            os.remove(temp_path)  # Cleanup temp files
        return " ".join(transcriptions)
    except Exception as e:
        return f"Error processing audio: {e}"

@spaces.GPU(duration=3)
def main():
    device = 0 if torch.cuda.is_available() else -1

    try:
        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

    def process_audio(audio_input):
        try:
            transcription = transcribe_long_audio(audio_input, transcriber, chunk_duration=30)
            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}", ""

    with gr.Blocks() as interface:
        with gr.Row():
            with gr.Column():
                # Enable recording or file upload
                audio_input = gr.Audio(type="numpy", label="Record or Upload Audio")
                process_button = gr.Button("Process Audio")
                stop_button = gr.Button("Stop Recording")  # Add Stop Button
            with gr.Column():
                transcription_output = gr.Textbox(label="Full Transcription", lines=10)
                summary_output = gr.Textbox(label="Summary", lines=5)

        def stop_microphone():
            """Dummy function to simulate stopping the microphone."""
            return "Recording stopped."

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

        stop_button.click(
            stop_microphone,
            inputs=[],
            outputs=[],
        )

    interface.launch(share=True)

if __name__ == "__main__":
    main()