app.py

import gradio as gr
import torch
from transformers import pipeline
import librosa
import soundfile as sf
import os
import spaces  # Ensure spaces is imported

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_path, transcriber, chunk_duration=30):
    """Transcribe long audio by splitting into smaller chunks."""
    try:
        # Load the audio file
        audio_data, sr = librosa.load(audio_path, sr=None)
        chunks = split_audio(audio_data, sr, chunk_duration)
        transcriptions = []
        for i, chunk in enumerate(chunks):
            chunk_path = f"temp_chunk_{i}.wav"
            sf.write(chunk_path, chunk, sr)  # Save chunk as WAV
            transcription = transcriber(chunk_path)["text"]
            transcriptions.append(transcription)
            os.remove(chunk_path)  # Cleanup temp files
        return " ".join(transcriptions)
    except Exception as e:
        print(f"Error in transcribe_long_audio: {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:
            print(f"Processing uploaded audio: {audio_input}")
            if not isinstance(audio_input, str):
                raise ValueError("Invalid input type. Please upload a valid audio file.")
            if os.path.isdir(audio_input):
                raise ValueError("Input is a directory, not a file.")

            # Transcribe the uploaded audio file
            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, audio_input
        except Exception as e:
            print(f"Error in process_audio: {e}")
            return f"Error processing audio: {e}", "", ""

    with gr.Blocks() as interface:
        with gr.Row():
            with gr.Column():
                # Only support file uploads
                audio_input = gr.Audio(type="filepath", label="Upload Audio File")
                process_button = gr.Button("Transcribe Audio")
            with gr.Column():
                transcription_output = gr.Textbox(label="Transcription", lines=10)
                summary_output = gr.Textbox(label="Summary", lines=5)

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

    interface.launch(share=False)

if __name__ == "__main__":
    main()