app.py

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

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

# ################ CHANGE THIS TO CHANGE THE LANGUAGE ###################### #
from TaiwaneseHokkien import TaiwaneseHokkien 

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

language = TaiwaneseHokkien(device=device, torch_dtype=torch_dtype)
# ########################################################################## #

# @spaces.GPU #[uncomment to use ZeroGPU]
def infer(
    audio,
    target 
):
    if type(audio) != tuple or type(target) != str: return [None, None]
    
    # preprocess 
    sampling_rate, wav = audio 
    if wav.ndim > 1:
        wav = wav.mean(axis=1)
    wav = wav.astype(np.float32)
    wav /= np.max(np.abs(wav))

    wav = librosa.resample(y=wav, orig_sr=sampling_rate, target_sr=16_000)
    
    user_pron = language.asr(wav)
    
    # compare texts 
    d_toks = language.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=language.compare_colors 
    )
    input_audio.input(fn=infer, inputs=[input_audio, target], outputs=[output, diff])
    
if __name__ == "__main__":
    demo.launch()