app.py

import os
import tempfile
from subprocess import Popen, PIPE
import torch
import gradio as gr
from transformers import pipeline
from transformers.pipelines.audio_utils import ffmpeg_read
from pydub import AudioSegment

# Constants
MODEL_NAME = "openai/whisper-large-v3-turbo"
BATCH_SIZE = 8
device = 0 if torch.cuda.is_available() else "cpu"

# Whisper pipeline
whisper_pipeline = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)

# Convert MP4 to MP3
def convert_mp4_to_mp3(mp4_path, mp3_path):
    try:
        audio = AudioSegment.from_file(mp4_path, format="mp4")
        audio.export(mp3_path, format="mp3")
    except Exception as e:
        raise RuntimeError(f"Error converting MP4 to MP3: {e}")

# Transcribe audio
def transcribe_audio(audio_path):
    try:
        # Ensure the audio file is read correctly in bytes format
        with open(audio_path, "rb") as audio_file:
            audio_data = audio_file.read()  # Read the file as bytes

        # Process the audio file for transcription
        inputs = ffmpeg_read(audio_data, whisper_pipeline.feature_extractor.sampling_rate)
        inputs = {"array": inputs, "sampling_rate": whisper_pipeline.feature_extractor.sampling_rate}

        # Run the transcription pipeline
        result = whisper_pipeline(inputs, batch_size=8, return_timestamps=False)
        return result["text"]
    except Exception as e:
        return f"Error during transcription: {e}"

# Gradio Interface Function
def transcribe_file(file):
    # Check file type and convert if necessary
    if file.name.endswith(".mp4"):
        temp_mp3_path = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False).name
        try:
            convert_mp4_to_mp3(file.name, temp_mp3_path)
            audio_path = temp_mp3_path
        except Exception as e:
            return f"Error during MP4 to MP3 conversion: {e}"
    else:
        audio_path = file.name

    # Transcribe audio
    transcription = transcribe_audio(audio_path)
    
    # Clean up temporary files
    if file.name.endswith(".mp4"):
        os.remove(temp_mp3_path)
    
    return transcription

# Gradio interface setup
def launch_gradio():
    with gr.Blocks() as demo:
        gr.Markdown("# Audio Transcription with Whisper Model")
        
        gr.Interface(
            fn=transcribe_file,
            inputs=gr.File(label="Upload Audio/Video File (MP4 or MP3)"),
            outputs=gr.Textbox(label="Transcribed Text"),
        )

    demo.launch(share=True)

# Run the Gradio app
if __name__ == "__main__":
    launch_gradio()