|
import os |
|
from extra import TTSTokenizer, VitsConfig, CharactersConfig, VitsCharacters |
|
import torch |
|
import numpy as np |
|
|
|
|
|
with open("chars.txt", 'r') as f: |
|
letters = f.read().strip('\n') |
|
model="en_female_vits_30hrs.pt" |
|
|
|
text = "My name is g p t, chat g p t" |
|
|
|
config = VitsConfig( |
|
text_cleaner="multilingual_cleaners", |
|
characters=CharactersConfig( |
|
characters_class=VitsCharacters, |
|
pad="<PAD>", |
|
eos="<EOS>", |
|
bos="<BOS>", |
|
blank="<BLNK>", |
|
characters=letters, |
|
punctuations="!¡'(),-.:;¿? ", |
|
phonemes=None) |
|
) |
|
tokenizer, config = TTSTokenizer.init_from_config(config) |
|
|
|
x = tokenizer.text_to_ids(text) |
|
x = torch.from_numpy(np.array(x)).unsqueeze(0) |
|
net = torch.jit.load(model) |
|
with torch.no_grad(): |
|
out2 = net(x) |
|
import soundfile as sf |
|
sf.write("jit.wav", out2.squeeze().cpu().numpy(), 22050) |
