app.py

import gradio as gr
import torch

# from IPython.display import Audio, display
from omegaconf import OmegaConf

torch.hub.download_url_to_file(
    "https://raw.githubusercontent.com/snakers4/silero-models/master/models.yml",
    "latest_silero_models.yml",
    progress=False,
)

all_models = OmegaConf.load("latest_silero_models.yml")

language="ru"
model_id = "v3_1_ru"
device = torch.device("cpu")

model, example_text = torch.hub.load(
    repo_or_dir="snakers4/silero-models",
    model="silero_tts",
    language=language,
    speaker=model_id,
)
model.to(device)  # gpu or cpu

sample_rate = 48000
speaker = "aidar"
put_accent = True
put_yo = True
example_text = "В недрах тундры выдры в г+етрах т+ырят в вёдра ядра к+едров."

models = list(all_models.tts_models.get(language).keys())

model, example_text = torch.hub.load(
    repo_or_dir='snakers4/silero-models',
    model='silero_tts',
    language='ru',
    speaker=model_id
)

def change_language(language):
    models = list(all_models.tts_models.get(language).keys())
    return model_input.update(choices=models)

def change_model(language, model_name):
    model, example_text = torch.hub.load(
        repo_or_dir='snakers4/silero-models',
        model='silero_tts',
        language=language,
        speaker=model_name
    )
    
    return speaker_input.update(choices=model.speakers)


def generate_audio_by_text(text, text_type, speaker):
    if text_type == 'SSML':
        return model.save_wav(
            ssml_text=text,
            speaker=speaker,
            sample_rate=sample_rate,
            put_accent=put_accent,
            put_yo=put_yo,
        )
    else:
        return model.save_wav(
            text=text,
            speaker=speaker,
            sample_rate=sample_rate,
            put_accent=put_accent,
            put_yo=put_yo,
        )

with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            language_input = gr.Dropdown(
                label="Language",
                choices=list(all_models.tts_models.keys()),
                value="ru",
                interactive=True,
            )

            model_input = gr.Dropdown(
                label="Model (based on selected language)",
                value="v3_1_ru",
                choices=models,
                interactive=True,
            )

            speaker_input = gr.Dropdown(
                label="Speaker (based on selected model)",
                value="kseniya",
                choices=model.speakers,
                interactive=True,
            )

            text_input = gr.Textbox(
                label="Text for generating",
                value="В недрах тундры выдры в г+етрах т+ырят в вёдра +ядра к+едров.",
                lines=5,
                interactive=True,
            )

            text_type_input = gr.Radio(
                label="Text type",
                choices=["Common", "SSML"],
                value="Common",
                interactive=True,
            )

            language_input.change(change_language, inputs=language_input, outputs=model_input)
            model_input.change(change_model, inputs=[language_input, model_input], outputs=speaker_input)

        with gr.Column():
            audio_output = gr.Audio(label="Output audio")
            generate_btn = gr.Button(value="Generate", variant="primary")
            generate_btn.click(
                generate_audio_by_text,
                inputs=[text_input, text_type_input, speaker_input],
                outputs=audio_output,
            )

    gr.Markdown(
        "This is a simple frontend for [silero](https://github.com/snakers4/silero-models) project (Text-To-Speech part only)."
    )
    gr.Markdown(
        "You can check [official docs](https://github.com/snakers4/silero-models/wiki/SSML) to find information about SSML syntax."
    )

demo.launch()