app.py

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

# 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 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:
        time.sleep(3)
        text = p(audio)["text"]
        state += text + "\n"
        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="filepath"),
        'state',
        gr.inputs.Audio(label="Upload Audio File", type="filepath", source="upload")
    ],
    outputs=[
        "textbox",
        "state"
    ],
    live=True).launch()