app.py

# coding=utf-8

import gradio as gr
import numpy as np
import soundfile as sf
import spaces
import torch
import torchaudio
from sv import process_audio


@spaces.GPU
def model_inference(input_wav, language):
    # Simplify language selection
    language = language if language else "auto"

    # Handle input_wav format
    if isinstance(input_wav, tuple):
        fs, input_wav = input_wav
        input_wav = input_wav.astype(np.float32) / np.iinfo(np.int16).max
        input_wav = input_wav.mean(-1) if len(input_wav.shape) > 1 else input_wav
        if fs != 16000:
            resampler = torchaudio.transforms.Resample(fs, 16000)
            input_wav = resampler(torch.from_numpy(input_wav).float()[None, :])[
                0
            ].numpy()

    # Process audio
    with sf.SoundFile("temp.wav", "w", samplerate=16000, channels=1) as f:
        f.write(input_wav)
    result = process_audio("temp.wav", language=language)

    return result


def launch():
    with gr.Blocks(theme=gr.themes.Soft()) as demo:
        with gr.Row():
            gr.Examples(
                examples=[["example/scb.mp3"]],
                inputs=[audio_inputs],
                outputs=text_outputs,
                fn=lambda x: model_inference(x, "yue"),
            )

        with gr.Row():
            with gr.Column(scale=2):
                audio_inputs = gr.Audio(label="Input")
                fn_button = gr.Button("Process Audio", variant="primary")

            with gr.Column(scale=3):
                text_outputs = gr.Textbox(lines=10, label="Output")

    demo.launch()


if __name__ == "__main__":
    launch()