app.py

import gradio as gr
from transformers import VitsModel, AutoTokenizer, set_seed
import torch
import scipy.io.wavfile
from ruaccent import RUAccent

speakers={"woman": 0,
        "man": 1}

model = VitsModel.from_pretrained("utrobinmv/tts_ru_free_hf_vits_low_multispeaker")
tokenizer = AutoTokenizer.from_pretrained("utrobinmv/tts_ru_free_hf_vits_low_multispeaker")
model.eval()
set_seed(555)

accentizer = RUAccent()
accentizer.load(omograph_model_size='turbo', use_dictionary=True)


def generate_audio(speaker_name, text):
    text = accentizer.process_all(text)
    inputs = tokenizer(text, return_tensors="pt")
    inputs['speaker_id'] = speakers[speaker_name]
    
    with torch.no_grad():
        output = model(**inputs).waveform
        
    scipy.io.wavfile.write("output.wav", rate=model.config.sampling_rate, data=output[0].cpu().numpy())
    
    return "output.wav"

speaker_dropdown = gr.Dropdown(
            choices=speakers,
            label="Speaker id",
            value='woman',
            info=f"Models are trained on 2 speakers",  
            interactive=True   
        )

iface = gr.Interface(fn=generate_audio, inputs=[speaker_dropdown,"text"], outputs="audio", title="Text to Speech Russian free multispeaker model", description="Введите текст на русском языке, чтобы преобразовать его в русскую звуковую речь. Пример текста: Привет, как дела? А у тебя как?")
iface.launch(share=True)