app.py

import gradio as gr
import time
from transformers import pipeline
import torch
import ffmpeg  # Make sure it's ffmpeg-python

# Check if GPU is available
use_gpu = torch.cuda.is_available()

# Configure the pipeline to use the GPU if available
if use_gpu:
    p = pipeline("automatic-speech-recognition",
                 model="carlosdanielhernandezmena/wav2vec2-large-xlsr-53-faroese-100h", device=0)
else:
    p = pipeline("automatic-speech-recognition",
                 model="carlosdanielhernandezmena/wav2vec2-large-xlsr-53-faroese-100h")

def extract_audio_from_m3u8(url):
    try:
        output_file = "output_audio.aac"
        ffmpeg.input(url).output(output_file).run(overwrite_output=True)
        return output_file
    except Exception as e:
        return f"An error occurred: {e}"

def transcribe_function(audio, state, m3u8_url):
    # If an m3u8 URL is provided, extract audio from the URL
    if m3u8_url:
        audio = extract_audio_from_m3u8(m3u8_url)

    # If no audio is provided, return the current state
    if not audio:
        return {state_var: state, transcription_var: state}

    try:
        time.sleep(3)  # Simulate processing delay
        text = p(audio, chunk_length_s=50)["text"]
        state += text + "\n"
        return {state_var: state, transcription_var: state}
    except Exception as e:
        return {transcription_var: "An error occurred during transcription.", state_var: state}

def reset_output(transcription, state):
    """Reset the state to an empty string."""
    return "", ""

with gr.Blocks() as demo:
    state_var = gr.State("")

    with gr.Row():
        with gr.Column():
            # Single Audio component supporting both recording and file upload
            audio_input = gr.Audio(type="filepath", label="Audio (Record or Upload)")
            m3u8_url = gr.Textbox(label="m3u8 URL | E.g.: from kvf.fo or logting.fo")
        with gr.Column():
            # Use interactive=False to make the textbox read-only
            transcription_var = gr.Textbox(type="text", label="Transcription", interactive=False)

    with gr.Row():
        transcribe_button = gr.Button("Transcribe")
        reset_button = gr.Button("Reset output")

    transcribe_button.click(
        transcribe_function,
        [audio_input, state_var, m3u8_url],
        [transcription_var, state_var]
    )

    reset_button.click(
        reset_output,
        [transcription_var, state_var],
        [transcription_var, state_var]
    )

demo.launch()