import os from extra import TTSTokenizer, VitsConfig, CharactersConfig, VitsCharacters import torch import numpy as np #ch female with open("chars.txt", 'r') as f: letters = f.read().strip('\n') model="mt_female_vits_30hrs.pt" # text = " হলেও আমাদের সবার সার্বিক শৃঙ্খলা বোধের উন্নতি হবে" text = "भेजना चाहते हैं हिंदी में मैसेज लेकिन नहीं आती टाइपिंग?" config = VitsConfig( text_cleaner="multilingual_cleaners", characters=CharactersConfig( characters_class=VitsCharacters, pad="", eos="", bos="", blank="", 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)