TTS.py

# ├── russian_g2p  # https://github.com/nsu-ai/russian_g2p.git
# ├── russian_stt_text_normalization # https://github.com/snakers4/russian_stt_text_normalization.git
# ├── # silero
# └── TTS.py
#
#

import itertools
import torch
import torchaudio
from russian_stt_text_normalization.normalizer import Normalizer
#from russian_g2p.russian_g2p.Transcription import Transcription
from russian_g2p.russian_g2p.Accentor import Accentor
your_accentor = Accentor()

language = 'ru'
gg=["baya_16khz", "aidar_16khz"]
speaker = gg[0] #'kseniya_16khz'
device = torch.device('cpu')
model, symbols, sample_rate, example_text, apply_tts = torch.hub.load(repo_or_dir='snakers4/silero-models',
                                                                      model='silero_tts',
                                                                      language=language,
                                                                      speaker=speaker)
model = model.to(device)  # gpu or cpu


import itertools
import torch
import torchaudio
from russian_stt_text_normalization.normalizer import Normalizer
#from russian_g2p.russian_g2p.Transcription import Transcription
from russian_g2p.russian_g2p.Accentor import Accentor
your_accentor = Accentor()

language = 'ru'
gg=["baya_16khz", "aidar_16khz"]
speaker = gg[0] #'kseniya_16khz'
device = torch.device('cpu')
model, symbols, sample_rate, example_text, apply_tts = torch.hub.load(repo_or_dir='snakers4/silero-models',
                                                                      model='silero_tts',
                                                                      language=language,
                                                                      speaker=speaker)
model = model.to(device)  # gpu or cpu


text = 'Принцесса была прекрасной'

norm = Normalizer()
result = norm.norm_text(text)
words  = [[w] for w in result.split() ]
accents = your_accentor.do_accents(words)
accents = " ".join(list(itertools.chain(*accents)))

print("1. ", text)
print("2. ", result)
print("3. ", accents)
audio = apply_tts(texts=[accents],
                  model=model,
                  sample_rate=sample_rate,
                  symbols=symbols,
                  device=device)

torchaudio.set_audio_backend("sox")
torchaudio.save('result.wav', audio[0], sample_rate)