app.py

import os

import gradio as gr
import numpy as np
import spaces
import torch
import TTS
from huggingface_hub import snapshot_download
from omegaconf import OmegaConf
from TTS.utils.synthesizer import Synthesizer

from ipa.ipa import get_ipa, parse_ipa
from replace.tts import ChangedVitsConfig

TTS.tts.configs.vits_config.VitsConfig = ChangedVitsConfig


def load_model(model_id):
    model_dir = snapshot_download(model_id)
    config_file_path = os.path.join(model_dir, "config.json")
    model_ckpt_path = os.path.join(model_dir, "model.pth")
    speaker_file_path = os.path.join(model_dir, "speakers.pth")
    language_file_path = os.path.join(model_dir, "language_ids.json")
    speaker_embedding_file_path = os.path.join(model_dir, "speaker_embs.pth")

    temp_config_path = "temp_config.json"
    with open(config_file_path, "r") as f:
        content = f.read()
        content = content.replace("speakers.pth", speaker_file_path)
        content = content.replace("language_ids.json", language_file_path)
        content = content.replace("speaker_embs.pth", speaker_embedding_file_path)
        f.close()
    with open(temp_config_path, "w") as f:
        f.write(content)
        f.close()

    return Synthesizer(
        tts_checkpoint=model_ckpt_path,
        tts_config_path=temp_config_path,
        use_cuda=torch.cuda.is_available(),
    )


OmegaConf.register_new_resolver("load_model", load_model)

models_config = OmegaConf.to_object(OmegaConf.load("configs/models.yaml"))


@spaces.GPU
def _do_tts(model_id, ipa, language_name, speaker_name=None, speaker_wav=None):
    model = models_config[model_id]["model"]
    if speaker_wav is not None:
        return model.tts(
            ipa,
            speaker_wav=speaker_wav,
            language_name=language_name,
            split_sentences=False,
        )
    return model.tts(
        ipa,
        speaker_name=speaker_name,
        language_name=language_name,
        split_sentences=False,
    )


def text_to_speech(
    model_id: str,
    use_default_emb_or_custom: str,
    speaker_wav,
    speaker: str,
    dialect,
    text: str,
):
    if len(text) == 0:
        raise gr.Error("請勿輸入空字串。")
    words, ipa, pinyin, missing_words = get_ipa(text, dialect=dialect)
    if len(missing_words) > 0:
        raise gr.Error(
            f"句子中的[{','.join(missing_words)}]目前無法轉成 ipa。請嘗試其他句子。"
        )
    parsed_ipa = parse_ipa(ipa)
    if use_default_emb_or_custom == "預設語者":
        wav = _do_tts(
            model_id,
            parsed_ipa,
            speaker_name=speaker
            if len(models_config[model_id]["speaker_mapping"]) > 1
            else None,
            language_name=dialect,
        )
    else:
        wav = _do_tts(
            model_id,
            parsed_ipa,
            speaker_wav=speaker_wav,
            language_name=dialect,
        )

    return (
        words,
        pinyin,
        (
            models_config[model_id]["model"].tts_model.config.audio.sample_rate,
            np.array(wav),
        ),
    )


def when_model_selected(model_id):
    model_config = models_config[model_id]

    speaker_drop_down_choices = [
        (k, v) for k, v in model_config["speaker_mapping"].items()
    ]

    dialect_drop_down_choices = [
        (k, v) for k, v in model_config["dialect_mapping"].items()
    ]

    use_default_emb_or_ref_radio_visible = False
    if model_config["model"].tts_model.config.model_args.speaker_encoder_model_path:
        use_default_emb_or_ref_radio_visible = True

    return (
        gr.update(
            choices=speaker_drop_down_choices,
            value=speaker_drop_down_choices[0][1]
            if len(speaker_drop_down_choices) > 0
            else None,
            interactive=len(speaker_drop_down_choices) > 1,
        ),
        gr.update(
            choices=dialect_drop_down_choices,
            value=dialect_drop_down_choices[0][1],
            interactive=len(dialect_drop_down_choices) > 1,
        ),
        gr.update(visible=use_default_emb_or_ref_radio_visible, value="預設語者"),
    )


def use_default_emb_or_custom_radio_input(use_default_emb_or_custom):
    if use_default_emb_or_custom == "客製化語者":
        return gr.update(visible=True), gr.update(visible=False)
    return gr.update(visible=False), gr.update(visible=True)


demo = gr.Blocks(
    title="臺灣客語語音生成系統",
    css="@import url(https://tauhu.tw/tauhu-oo.css);",
    theme=gr.themes.Default(
        font=(
            "tauhu-oo",
            gr.themes.GoogleFont("Source Sans Pro"),
            "ui-sans-serif",
            "system-ui",
            "sans-serif",
        )
    ),
)

with demo:
    default_model_id = list(models_config.keys())[0]
    model_drop_down = gr.Dropdown(
        models_config.keys(),
        value=default_model_id,
        label="模型",
    )
    use_default_emb_or_custom_radio = gr.Radio(
        label=None,
        choices=["預設語者", "客製化語者"],
        value="預設語者",
        visible=True,
        show_label=False,
    )
    speaker_wav = gr.Audio(
        label="客製化語音",
        visible=False,
        editable=False,
        type="filepath",
        waveform_options=gr.WaveformOptions(
            show_controls=False,
            sample_rate=16000,
        ),
    )
    speaker_drop_down = gr.Dropdown(
        choices=[
            (k, v)
            for k, v in models_config[default_model_id]["speaker_mapping"].items()
        ],
        value=list(models_config[default_model_id]["speaker_mapping"].values())[0],
        label="語者",
        interactive=len(models_config[default_model_id]["speaker_mapping"]) > 1,
    )
    use_default_emb_or_custom_radio.input(
        use_default_emb_or_custom_radio_input,
        inputs=[use_default_emb_or_custom_radio],
        outputs=[speaker_wav, speaker_drop_down],
    )

    dialect_drop_down = gr.Dropdown(
        choices=[
            (k, v)
            for k, v in models_config[default_model_id]["dialect_mapping"].items()
        ],
        value=list(models_config[default_model_id]["dialect_mapping"].values())[0],
        label="腔調",
        interactive=len(models_config[default_model_id]["dialect_mapping"]) > 1,
    )

    model_drop_down.input(
        when_model_selected,
        inputs=[model_drop_down],
        outputs=[speaker_drop_down, dialect_drop_down, use_default_emb_or_custom_radio],
    )

    gr.Markdown(
        """
        # 臺灣客語語音合成系統
        ### Taiwanese Hakka Text-to-Speech System
        ### 研發團隊
        - **[李鴻欣 Hung-Shin Lee](mailto:[email protected])（[聯和科創](https://www.104.com.tw/company/1a2x6bmu75)）**
        - **[陳力瑋 Li-Wei Chen](mailto:[email protected])（[聯和科創](https://www.104.com.tw/company/1a2x6bmu75)）**
        ### 合作單位
        - **[國立聯合大學智慧客家實驗室](https://www.gohakka.org)**
        """
    )
    gr.Interface(
        text_to_speech,
        inputs=[
            model_drop_down,
            use_default_emb_or_custom_radio,
            speaker_wav,
            speaker_drop_down,
            dialect_drop_down,
            gr.Textbox(
                label="輸入文字",
                value="客家族群个六堆運動會會一直延續下去，為臺灣个體育史寫下特別个一頁。",
            ),
        ],
        outputs=[
            gr.Textbox(interactive=False, label="斷詞"),
            gr.Textbox(interactive=False, label="客語拼音"),
            gr.Audio(interactive=False, label="合成語音", show_download_button=True),
        ],
        allow_flagging="auto",
    )

demo.launch()