app.py

import os

os.environ['CUDA_VISIBLE_DEVICES'] = ''
os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true'

import malaya_speech
from malaya_speech.utils.astype import float_to_int
import gradio as gr
import logging
import json

logging.basicConfig(level=logging.INFO)

MODELS = {}
AVAILABLE_MODELS = malaya_speech.tts.available_vits().index.tolist()

def tts(text, model, temperature, length_ratio):
    global MODELS
    if model not in MODELS:
        MODELS[model] = malaya_speech.tts.vits(model=model)
    
    r = MODELS[model].predict(
        text, 
        temperature = temperature,
        length_ratio = length_ratio,
    )
    data = float_to_int(r['y'])
    return (22050, data)

demo = gr.Interface(
    fn=tts,
    inputs=[
        gr.components.Textbox(label='Text'),
        gr.components.Dropdown(label='Available models', choices=AVAILABLE_MODELS, value = 'mesolitica/VITS-osman'),
        gr.Slider(0.0, 1.0, value=0.6666, label='temperature, changing this will manipulate pitch'),
        gr.Slider(0.0, 3.0, value=1.0, label='length ratio, changing this will manipulate duration output'),
    ],
    outputs=['audio'],
    examples=[
        ['Syed Saddiq berkata, mereka seharusnya mengingati bahawa semasa menjadi Perdana Menteri Pakatan Harapan', 'mesolitica/VITS-osman', 0.6666, 1.0],
        ['SHAH ALAM - Pertubuhan Kebajikan Anak Bersatu Selangor bersetuju pihak kerajaan mewujudkan Suruhanjaya Siasatan Diraja untuk menyiasat isu kartel daging.', 'mesolitica/VITS-haqkiem', 0.6666, 1.0],
    ],
    cache_examples=False,
    title='End-to-End TTS using VITS from Mesolitica',
)

demo.launch(server_name='0.0.0.0')