app.py

import gradio as gr
import time
import torch
from transformers import pipeline
import numpy as np

# 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")

chunk_size = 30  # Adjust the chunk size as needed

def transcribe(audio, state="", uploaded_audio=None):
    if uploaded_audio is not None:
        audio = uploaded_audio
    if not audio:
        return state, state  # Return a meaningful message

    try:
        state += "Transcribing...\n"
        chunks = [audio[i:i + chunk_size] for i in range(0, len(audio), chunk_size)]
        for chunk in chunks:
            text = p(chunk)["text"]
            state += text + "\n"
            time.sleep(1)  # Simulate processing time for each chunk
        return state, state
    except Exception as e:
        return "An error occurred during transcription.", state  # Handle other exceptions

gr.Interface(
    fn=transcribe, 
    inputs=[
        gr.inputs.Audio(source="microphone", type="numpy"),
        'state',
        gr.inputs.Audio(label="Upload Audio File", type="numpy", source="upload")
    ],
    outputs=[
        "textbox",
        "state"
    ],
    live=True
).launch()