app.py - main - diger/fb2tts

app.py
282 строки9 KB
diger
fixed debug
03 июн 2026, 14:06
03 июн 2026, 14:063401eac
import gradio as gr
import shutil
import logging
import argparse
from pathlib import Path
from config import config
from libs.utils import get_data_list, now_dir, change_pitch
from libs.accent import accentizer
from libs.tts import synth
from libs.tts_preprocessor import TextParse
from gr_tabs.parse_tab import parse_tab
from gr_tabs.tts_tab import tts_tab
from gr_tabs.settings_tab import settings_tab
from gr_tabs.cover_tab import cover_tab

gr.set_static_paths(paths=[Path(config.data_path),])
txt_parser = TextParse(False)

def refresh_data(ab_name):
    return {"value": ab_name, "choices": sorted(get_data_list()), "__type__": "update"}

def remove_dataset(ab_name):
    ab_path = Path(config.data_path) / ab_name
    gr.Warning(f'Удаляем {ab_name}')
    
    try:
        if ab_path.exists():
            shutil.rmtree(ab_path)
    except Exception as e:
        gr.Warning(f"Ошибка при удалении: {e}")
    
    return {"value": '', "choices": sorted(get_data_list()), "__type__": "update"}, \
            {"visible": False, "__type__": "update"}

def save_drop_dataset_audio(dropbox):
    dropbox_path = Path(dropbox)
    ab_file = dropbox_path.name
    ab_name = dropbox_path.stem
    ab_path = Path(config.data_path) / ab_name
    
    if not ab_path.exists():
        ab_path.mkdir(parents=True, exist_ok=True)
    
    destination_path = ab_path / ab_file
    if destination_path.exists():
        destination_path.unlink()
    
    try:
        content = dropbox_path.read_bytes()
        destination_path.write_bytes(content)
        gr.Info("Файл загружен", duration=4)
    except Exception as e:
        gr.Warning(f"Ошибка при копировании файла: {e}")
        return refresh_data('')
    
    return refresh_data(ab_name)

def toggle_tab(ab_path):
    return gr.Tabs(visible=True, selected=0), ab_path

def put_accents(string):
    string = accentizer.process_accent(string, r'\\+\\w+|\\w+\\+\\w+')
    return string

def text_to_audio(string, spk, rate=1, noise=None, pitch=None, ref_audio=None, ref_text=''):
    if not synth:
        return gr.update(label = "Модель не загружена!")
    np_audio, sr = synth.synth_audio(txt_parser.garbage(string), spk, rate, noise, ref_audio, ref_text)
    if pitch != 50:
        np_audio = change_pitch(np_audio, sr, pitch)
    return (sr, np_audio)

tts_models_list = [
    ('Vosk 0.9 (5 голосов)', 1),
    ('Vosk 0.10 (dev. 56 голосов)', 2),
    ('Silero v5 (5 голосов)', 3),
    ('Silero v5_cis (60 голосов)', 4),
    ('Misha24-10 (F5-TTS)', 5),
    ('ESpeech-TTS (F5-TTS)', 6)
]

accent_models_list = [
    ('RuAccent', 1),
    ('Silero stress', 2),
]

def tts_model_load(ver, progress=gr.Progress()):
    return synth.load(ver)

def acc_model_load(ver, progress=gr.Progress()):
    return accentizer.load(ver)

def change_tts_model(mver):
    sp_list = synth.speakers_list()
    speaker = sp_list[0]
    if isinstance(speaker, tuple):
        speaker = sp_list[0][1]
    if mver == 5 or mver ==6:
        ref_group = gr.update(visible=True)
    else:
        ref_group = gr.update(visible=False)
    return (
        gr.update(value=speaker, choices=sp_list),
        gr.update(visible=True),
        gr.update(interactive=True),
        ref_group
    )

def change_acc_model(mver):
    return gr.update(interactive=True)

custom_theme = gr.themes.ThemeClass.load("themes/argilla.json")

with gr.Blocks(title="FB2 to speech 🇷🇺") as App:

    tts_state = gr.State()
    acc_state = gr.State()
    with gr.Sidebar():
        tts_sel = gr.Dropdown(
            value='',
            allow_custom_value=True,
            label='Выбрать модель TTS',
            choices=tts_models_list,
            interactive=True,
        )
        tts_status = gr.Textbox(show_label=False, visible=True)
        acc_sel = gr.Dropdown(
            value='',
            allow_custom_value=True,
            label='Расстановка ударений',
            choices=accent_models_list,
            interactive=True,
        )
        acc_status = gr.Textbox(show_label=False, visible=True)

        tts_sel.select(
            tts_model_load,
            inputs=tts_sel,
            outputs=[tts_state, tts_status],
            show_progress_on=tts_status
        )
        acc_sel.select(
            acc_model_load,
            inputs=acc_sel,
            outputs=[acc_state, acc_status],
            show_progress_on=acc_status
        )
    with gr.Tabs():
        with gr.Tab(label="Demo TTS"):
            with gr.Group(visible=False) as f5_gr:
                with gr.Row():
                    ref_audio = gr.Audio(label='Ваш образец голоса')
                with gr.Row():
                    ref_text = gr.Textbox(
                        label='Текст в образце',
                        lines=1,
                        placeholder="Введите текст, который звучит в образце",
                        interactive=True
                    )
            with gr.Row():
                spk_sel = gr.Dropdown(
                    value='',
                    label='Выбрать голос',
                    choices=[''],
                    interactive=True,
                )
                speech_rate = gr.Slider(
                    0,
                    3,
                    1,
                    step=0.1,
                    label="Задать скорость",
                    interactive=True,
                )
                noise_lvl = gr.Slider(
                    0,
                    64,
                    16,
                    step=1,
                    label="Уровень шума(выше-лучше)",
                    interactive=True,
                )
                pitch_sel = gr.Slider(
                    0,
                    100,
                    50,
                    step=1,
                    label="Тембр",
                    interactive=True,
                )
            with gr.Row():
                text_input = gr.Textbox(label='Текст', lines=2, placeholder="Введите текст для озвучки", interactive=True, max_length=220)
                audio_output = gr.Audio(interactive=False, buttons=[])
            with gr.Row():
                accent_button = gr.Button("Проставить ударения", interactive=False)
                tts_button = gr.Button("Преобразовать в речь", interactive=False)

            accent_button.click(
                put_accents,
                inputs=text_input,
                outputs=text_input
            )
            tts_button.click(
                text_to_audio,
                inputs=[text_input, spk_sel, speech_rate, noise_lvl, pitch_sel,ref_audio,ref_text],
                outputs=audio_output
            )
        with gr.Tab("Fb2TTS", visible=False, id="fb2tts") as fb2tts_tab:
            ab_state = gr.State()
            with gr.Row():
                ab_path = gr.Dropdown(
                    value='',
                    show_label=False,
                    allow_custom_value=True,
                    choices=get_data_list(),
                    interactive=True,
                )
                rm_dataset = gr.Button("❌ Удалить АК")
                upload_text_file = gr.UploadButton("Загрузить файл", file_count="single", file_types=[".fb2", ".txt"])
            with gr.Tabs(visible=False) as tabs:
                cover_tab(ab_path, ab_state)
                parse_tab(ab_path, acc_state, tts_state)
                tts_tab(ab_path, tts_state)

            upload_text_file.upload(
                fn=save_drop_dataset_audio,
                inputs=upload_text_file,
                outputs=[ab_path],
            ).then(
                toggle_tab,
                inputs=ab_path,
                outputs=[tabs, ab_state]
            )

            rm_dataset.click(
                remove_dataset,
                inputs=ab_path,
                outputs=[ab_path, tabs]
            )

            ab_path.select(
                toggle_tab,
                inputs=ab_path,
                outputs=[tabs, ab_state]
            )

        with gr.Tab("Настройки"):
            settings_tab(tts_state)

    tts_state.change(
        change_tts_model,
        inputs=tts_state,
        outputs=[spk_sel, fb2tts_tab, tts_button, f5_gr]
    )

    acc_state.change(
        change_acc_model,
        inputs=acc_state,
        outputs=accent_button
    )

if __name__ == "__main__":
    sound_dir = now_dir / "sound"
    
    parser = argparse.ArgumentParser(description="Параметры запуска TTS-сервера")
    parser.add_argument("--port", type=int, default=None, help="Порт для запуска сервера")
    parser.add_argument("--share", action="store_true", default=None, help="Создать публичную ссылку")
    parser.add_argument("--debug", action="store_true", default=None, help="Режим отладки")
    parser.add_argument("--server-name", type=str, default=None, help="Адрес сервера")
    args = parser.parse_args()

    config.debug = args.debug
    
    App.launch(
        theme=custom_theme,
        server_name=args.server_name or config.server_name,
        server_port=args.port or config.port,
        share=args.share or config.share,
        debug=args.debug or config.debug,
        allowed_paths=[str(sound_dir)],
        favicon_path='libs/fb2tts.ico'
    )