app.py

from pathlib import Path
from vietTTS.hifigan.mel2wave import mel2wave
from vietTTS.nat.text2mel import text2mel
from vietTTS import nat_normalize_text
import numpy as np

import os
os.system("gdown --id 16UhN8QBxG1YYwUh8smdEeVnKo9qZhvZj -O duration_latest_ckpt.pickle")
os.system("gdown --id 1-8Ig65S3irNHSzcskT37SLgeyuUhjKdj -O acoustic_latest_ckpt.pickle")
os.system("gdown --id 19cRNDC6IrHFAAE4U9I7K0mzLMgPsi5zb -O hk_hifi.pickle")


def text_to_speech(text):
    text = nat_normalize_text(text)
    mel = text2mel(
      text, 
      "lexicon.txt", 
      0.2, 
      Path("acoustic_latest_ckpt.pickle"), 
      Path("duration_latest_ckpt.pickle")
    )
    wave = mel2wave(mel, Path("config.json"), Path("hk_hifi.pickle"))
    return (wave * (2**15)).astype(np.int16)
    import gradio as gr


def speak(text):
    y = text_to_speech(text)
    return 16_000, y

title = "vietTTS"
description = "A vietnamese text-to-speech demo."

iface = gr.Interface(
    fn=speak, 
    inputs="text", 
    outputs="audio",
    title = title,
    description=description
)

iface.launch()