app.py

import gradio as gr
import numpy as np
from difflib import Differ

# import spaces #[uncomment to use ZeroGPU]
import torch

from transformers import pipeline 

device = "cuda" if torch.cuda.is_available() else "cpu"
model_repo_id = "emlinking/wav2vec2-large-xls-r-300m-tsm-asr-v6" 

if torch.cuda.is_available():
    torch_dtype = torch.float16
else:
    torch_dtype = torch.float32

pipe = pipeline(task="automatic-speech-recognition", model=model_repo_id, device=device)

# @spaces.GPU #[uncomment to use ZeroGPU]
def infer(
    audio,
    target 
):
    sampling_rate, wav = audio 
    if wav.ndim > 1:
        wav = wav.mean(axis=1)
    wav = wav.astype(np.float32)
    wav /= np.max(np.abs(wav))
    user_pron = pipe(wav)['text']
    
    # compare texts 
    d = Differ() 
    d_toks = [(i[2:], i[0] if i[0] != " " else None) for i in d.compare(target, user_pron)]
    return (user_pron, d_toks)

css = """
#col-container {
    margin: 0 auto;
    max-width: 640px;
}
"""

with gr.Blocks(css=css) as demo:
    gr.Markdown(" # PhonoLearn")
    target = gr.Textbox(label='Practice Sentence (Tâi-lô)')
    input_audio = gr.Audio(
        sources=["microphone", "upload"]
    )
    output = gr.Textbox(label='Your Pronunciation')
    diff = gr.HighlightedText(
        label='Comparison',
        combine_adjacent=True,
        show_legend=True,
        color_map={'+': 'red', '-': 'green'}
    )
    input_audio.input(fn=infer, inputs=[input_audio, target], outputs=[output, diff])
    
if __name__ == "__main__":
    demo.launch()