app.py

import torch

from transformers import pipeline

import numpy as np
import gradio as gr

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

pipe = pipeline("text-to-speech", model="kakao-enterprise/vits-ljs", device=device)

# Inference
def generate_audio(text):

    output = pipe(text)
    output =  gr.Audio(value = (output["sampling_rate"], output["audio"].squeeze()), type="numpy", autoplay=False, label="Response Voice Player", show_label=True,
                               visible=True)
    
    ###############language = "english"
    return output

css = """
#container{
    margin: 0 auto;
    max-width: 80rem;
}
#intro{
    max-width: 100%;
    text-align: center;
    margin: 0 auto;
}
"""

# Gradio blocks demo    
with gr.Blocks(css=css) as demo_blocks:

    with gr.Row():
        with gr.Column():
            inp_text = gr.Textbox(label="Input Text", info="What sentence would you like to synthesise?")
            btn = gr.Button("Generate Audio!")

            
        with gr.Column():
            out_audio = gr.Audio(type="numpy", autoplay=True, label="Generated Audio - British Female Speaker", show_label=True, visible=True)

    btn.click(generate_audio, [inp_text], out_audio)
    

demo_blocks.queue().launch()