app.py

import torch

from transformers import pipeline

import numpy as np
import gradio as gr

def _grab_best_device(use_gpu=True):
    if torch.cuda.device_count() > 0 and use_gpu:
        device = "cuda"
    else:
        device = "cpu"
    return device

device = _grab_best_device()

HUB_PATH = "ylacombe/vits_vctk_welsh_male"
pipe = pipeline("text-to-speech", model=HUB_PATH, device=0)

title = "# 🐶 VITS"

description = """

"""

num_speakers = pipe.model.config.num_speakers

# Inference
def generate_audio(text):

    out = []
    for i in range(num_speakers):
        forward_params = {"speaker_id": i}
        output = pipe(text, forward_params=forward_params)
        out.append((output["sampling_rate"], output["audio"].squeeze()))
    
    return out


# Gradio blocks demo    
with gr.Blocks() as demo_blocks:
    gr.Markdown(title)
    gr.Markdown(description)
    with gr.Row():
        with gr.Column():
            inp_text = gr.Textbox(label="Input Text", info="What would you like bark to synthesise?")
            btn = gr.Button("Generate Audio!")
    
        with gr.Column():
            outputs = []
            for i in range(num_speakers):
                out_audio = gr.Audio(type="numpy", autoplay=False, label=f"Generated Audio {i}", show_label=True)
                outputs.append(out_audio)
    
    btn.click(generate_audio, [inp_text], [outputs])

demo_blocks.launch()