app.py

import time
import os
import logging

import gradio as gr
import numpy as np
import pandas as pd
from pypinyin import lazy_pinyin
from i18n import gettext, Translate

from api import generate_api, get_audio, generate_voice, load_characters_csv
from utils import get_length

# 翻译文件位置
trans_file = os.path.join(os.path.dirname(__file__), "i18n", "translations.json")

# 关闭aiohttp的DEBUG日志
logging.getLogger("aiohttp").setLevel(logging.WARNING)
# logging.getLogger("gradio").setLevel(logging.WARNING)

# 带有时间的log
logging.basicConfig(
    level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)


header = """header"""

terms = "terms"


def update_all_characters(lang, category):
    new_characters, category = load_characters_csv(lang)
    initial_characters = get_characters(kind=category[0], all_characters=new_characters)
    return (
        new_characters,
        initial_characters,
        gr.Gallery(
            value=[[char["头像"], char["名称"]] for char in initial_characters],
            show_label=False,
            elem_id="character_gallery",
            columns=[11],
            object_fit="contain",
            height="auto",
            interactive=False,
            allow_preview=False,
            selected_index=None,
        ),
        category,
        gr.update(choices=category, value=category[0])
    )


def get_characters(
    query=None, page=1, per_page=400, kind="原神", lang="zh", all_characters=None
):
    # 使用传入的 all_characters 参数
    filtered_characters = all_characters[all_characters["类别"] == kind]

    if query:
        # 使用拼音和汉字进行搜索
        filtered_characters = filtered_characters[
            filtered_characters["名称"].str.contains(query, case=False)
        ]
        if filtered_characters.empty and lang == "zh":
            filtered_characters = all_characters[all_characters["类别"] == kind]
            filtered_characters = filtered_characters[
                filtered_characters["名称"]
                .apply(lambda x: "".join(lazy_pinyin(x)))
                .str.contains(query, case=False)
            ]

    # 按名称分组，并选择每组的第一个记录
    unique_characters = (
        filtered_characters.groupby("名称").first().reset_index().sort_values(by="id")
    )

    # 处理头像数据
    import pickle
    
    def process_avatar(avatar):
        if not isinstance(avatar, str):
            try:
                return pickle.loads(bytes(avatar))
            except:
                return avatar
        return avatar
    
    unique_characters['头像'] = unique_characters['头像'].apply(process_avatar)

    # 应用分页
    start_index = (page - 1) * per_page
    end_index = start_index + per_page

    return unique_characters.iloc[start_index:end_index].to_dict("records")


async def generate(selected_character=None, selected_characters=[], text="", lang="zh"):
    if selected_character:
        characters = [selected_character] + selected_characters
    else:
        characters = selected_characters
    if not selected_character and not selected_characters:
        if lang == "zh":
            raise gr.Error("请先选择一个角色")
        elif lang == "en":
            raise gr.Error("Please select a character first")
        elif lang == "ja":
            raise gr.Error("まず、キャラクターを選択してください")
        elif lang == "ko":
            raise gr.Error("먼저 캐릭터를 선택하세요")
    voice_ids = [char.get("voice_id") for char in characters if char.get("voice_id")]

    if not voice_ids:
        raise gr.Error("所选角色没有关联的 voice_id")

    start_time = time.time()
    # 假设我们只使用第一个选择的角色的名称
    if voice_ids == "1":
        if lang == "zh":
            raise gr.Error("该角色暂未创建语音")
        elif lang == "en":
            raise gr.Error("The character has not been created yet")
        elif lang == "ja":
            raise gr.Error("そのキャラクターの音声はまだ作成されていません")
        elif lang == "ko":
            raise gr.Error("해당 캐릭터의 음성이 아직 생성되지 않았습니다")

    if text == "":
        if lang == "zh":
            raise gr.Error("请输入需要合成的文本")
        elif lang == "en":
            raise gr.Error("Please enter the text to be synthesized")
        elif lang == "ja":
            raise gr.Error("合成するテキストを入力してください")
        elif lang == "ko":
            raise gr.Error("합성할 텍스트를 입력하세요")

    if get_length(text) > 1024:
        if lang == "zh":
            raise gr.Error("长度请控制在1024个字符以内")
        elif lang == "en":
            raise gr.Error("The text length exceeds 1024 words")
        elif lang == "ja":
            raise gr.Error("テキストの長さが1024文字を超えています")
        elif lang == "ko":
            raise gr.Error("텍스트 길이가 1024자를 초과합니다")

    audio = await generate_api(voice_ids, text)
    end_time = time.time()
    if lang == "zh":
        cost_time = f"合成共花费{end_time - start_time:.2f}秒"
    elif lang == "en":
        cost_time = (
            f"Total time spent synthesizing: {end_time - start_time:.2f} seconds"
        )
    elif lang == "ja":
        cost_time = f"合成にかかった時間: {end_time - start_time:.2f}秒"
    elif lang == "ko":
        cost_time = f"합성에 소요된 시간: {end_time - start_time:.2f}초"
    if isinstance(audio, str):
        print(audio)
        raise gr.Error(audio)
    else:
        return audio, cost_time


def get_character_emotions(character, all_characters):
    # 从all_characters中筛选出与当前角色名称相同的所有记录
    character_records = all_characters[all_characters["名称"] == character["名称"]]

    # 按情绪去重并获取完整的角色信息
    character_infos = character_records.drop_duplicates(subset=["情绪"]).to_dict(
        "records"
    )

    # 如果没有找到角色信息，返回一个包含默认值的字典
    return (
        character_infos
        if character_infos
        else [{"名称": character["名称"], "情绪": "默认情绪"}]
    )


def update_character_info(character_name, emotion, current_character, all_characters):
    character_info = None
    if character_name and emotion:
        character_info = all_characters[
            (all_characters["名称"] == character_name)
            & (all_characters["情绪"] == emotion)
        ]
    if character_name == "":
        return None
    character_info = character_info.iloc[0].to_dict()
    return character_info, all_characters


def add_new_voice(current_character, selected_characters, kind, lang, all_characters):
    if not current_character:
        if lang == "zh":
            raise gr.Error("请先选择一个角色")
        elif lang == "en":
            raise gr.Error("Please select a character first")
        elif lang == "ja":
            raise gr.Error("まず、キャラクターを選択してください")
        elif lang == "ko":
            raise gr.Error("먼저 캐릭터를 선택하세요")

    if len(selected_characters) >= 5:
        raise gr.Error("已达到最大选择数（5个）")

    # 检查是否已存在相同角色
    existing_char = next(
        (
            char
            for char in selected_characters
            if char["名称"] == current_character["名称"]
        ),
        None,
    )
    if existing_char:
        # 如果情绪不同,更新情绪
        if existing_char["情绪"] != current_character["情绪"]:
            existing_char["情绪"] = current_character["情绪"]
    else:
        selected_characters.insert(0, current_character)

    updated_characters = get_characters(
        kind=kind, lang=lang, all_characters=all_characters
    )
    # ! 取消gallery选中状态,返回个新的gallery是必要的，否则会保留上一次的选中状态。这里sonnet很喜欢改成返回一个数组，但这不能清空gallery的选中状态
    updated_gallery = gr.Gallery(
        value=[[char["头像"], char["名称"]] for char in updated_characters],
        show_label=False,
        elem_id="character_gallery",
        columns=[11],
        object_fit="contain",
        height="auto",
        interactive=False,
        allow_preview=False,
        selected_index=None,
    )

    return (
        None,
        gr.update(value=""),
        gr.update(choices=[]),
        selected_characters,
        updated_characters,
        updated_gallery,
        gr.update(visible=True),
        all_characters,
    )


def update_selected_chars_display(selected_characters):
    updates = []
    for i, (name, emotion, _, row) in enumerate(selected_chars_rows):
        if i < len(selected_characters):
            char = selected_characters[i]
            updates.extend(
                [
                    gr.update(value=char["名称"], visible=True),
                    gr.update(value=char["情绪"], visible=True),
                    gr.update(visible=True),
                    gr.update(visible=True),
                ]
            )
        else:
            updates.extend(
                [
                    gr.update(value="", visible=False),
                    gr.update(value="", visible=False),
                    gr.update(visible=False),
                    gr.update(visible=False),
                ]
            )
    return updates


def remove_character(index, selected_characters):
    if 0 <= index < len(selected_characters):
        del selected_characters[index]
    return selected_characters, gr.update(visible=True)


def update_gallery(kind, query, all_characters):
    updated_characters = get_characters(
        kind=kind, query=query, lang=lang, all_characters=all_characters
    )
    return (
        updated_characters,
        [[char["头像"], char["名称"]] for char in updated_characters],
        all_characters,
    )


def on_select(evt: gr.SelectData, characters, selected_characters, all_characters):
    # 如果没有选择角色，换人的时候清空
    if len(selected_characters) == 0:
        selected_characters = []

    selected = characters[evt.index]
    emotions = get_character_emotions(selected, all_characters)
    normal_index = 0
    for index, emotion in enumerate(emotions):
        if (
            emotion["情绪"] == "正常"
            or emotion["情绪"] == "보통"
            or emotion["情绪"] == "normal"
        ):
            normal_index = index
            break

    default_emotion = emotions[normal_index]["情绪"] if emotions else ""
    default_voice_id = emotions[normal_index]["voice_id"] if emotions else ""

    character_dict = selected.copy()
    character_dict["情绪"] = default_emotion
    character_dict["voice_id"] = default_voice_id
    return (
        selected["名称"],
        gr.Dropdown(
            choices=[emotion["情绪"] for emotion in emotions], value=default_emotion
        ),
        character_dict,
        selected_characters,
    )


async def update_prompt_audio(current_character):
    if current_character:
        return await get_audio(current_character.get("voice_id"))
    else:
        return None

async def create_voice(avatar, name, emotion, tags, gender, audio_data, lang):
    updates = {}
    for field, value in [("avatar", avatar), ("name", name), ("emotion", emotion), ("tags", tags), ("gender", gender), ("audio_data", audio_data)]:
        if field in ["avatar", "audio_data"]:
            if value is None or (isinstance(value, np.ndarray) and value.size == 0):
                updates[field] = gr.update(value=None)
        elif value == "":
            updates[field] = gr.update(value="")
    
    if updates:
        if lang == "zh":
            gr.Warning("请填写完整信息")
        elif lang == "en":
            gr.Warning("Please fill in all the information")
        elif lang == "ja":
            gr.Warning("すべての情報を入力してください")
        elif lang == "ko":
            gr.Warning("모든 정보를 입력하세요")
        return tuple(updates.get(field, gr.update()) for field in ["avatar", "name", "emotion", "tags", "gender", "audio_data"])
    duration = len(audio_data[1]) / audio_data[0]
    if duration < 3.2 or duration > 8:
        if lang == "zh":
            gr.Warning("音频时长请控制在3.2-8秒之间")
        elif lang == "en":
            gr.Warning("The audio duration should be between 3.2 and 8 seconds")
        elif lang == "ja":
            gr.Warning("音声の長さは3.2秒から8秒の間にしてください")
        elif lang == "ko":
            gr.Warning("음성 길이는 3.2초에서 8초 사이로 설정해야 합니다")
        return avatar, name, emotion, tags, gender, audio_data
    await generate_voice(avatar, name, emotion, tags, gender, audio_data, lang)
    if lang == "zh":
        gr.Info("创建成功")
    elif lang == "en":
        gr.Info("Create successfully")
    elif lang == "ja":
        gr.Info("作成に成功しました")
    elif lang == "ko":
        gr.Info("생성 성공")
    return avatar, name, emotion, tags, gender, audio_data

head = """
<title>Free Online Text to Speech (TTS) | Convert Text to Audio</title>
<meta name="description" content="Text to Speech(TTS) for free! 5-second voice cloning, no sign-up required.">
<meta name="keywords" content="text to speech, TTS, free TTS, online TTS, speech synthesis, voice generator">
"""
with gr.Blocks(title="Online Free TTS", theme=gr.themes.Soft(), head=head) as demo:
    gr.Markdown(
        "Online Free TTS(Text-to-Speech). Ultra-low latency, 5-second voice cloning."
    )
    lang = gr.Radio(
        choices=[("中文", "zh"), ("English", "en"), ("日本語", "ja"), ("한국인", "ko")],
        label=gettext("Language"),
        value="en",
        scale=1,
    )
    all_characters_state = gr.State(load_characters_csv("en")[0])
    category = gr.State(load_characters_csv("en")[1])
    
    with Translate(trans_file, lang, placeholder_langs=["en", "zh", "ja", "ko"]):
        gr.Markdown(value=gettext(header))
        with gr.Group():
            initial_characters = get_characters(
                kind="原神", lang="zh", all_characters=all_characters_state.value
            )
            characters = gr.State(initial_characters)
            selected_characters = gr.State([])
            current_character = gr.State(None)

        with gr.Tab(gettext("Synthesis Voice")):
            with gr.Blocks():
                with gr.Row():
                    kind = gr.Dropdown(
                        choices=category.value,
                        value="原神",
                        label=gettext("Select character category"),
                    )
                    query = gr.Textbox(
                        label=gettext("Search character"),
                        value="",
                        lines=1,
                        max_lines=1,
                        interactive=True,
                    )
                with gr.Blocks():
                    gallery = gr.Gallery(
                        value=[
                            [char["头像"], char["名称"]] for char in characters.value
                        ],
                        show_label=False,
                        elem_id="character_gallery",
                        columns=[11],
                        object_fit="contain",
                        height="auto",
                        interactive=False,
                        allow_preview=False,
                        selected_index=None,
                    )
                with gr.Row():
                    character_name = gr.Textbox(
                        label=gettext("Currently selected character"),
                        interactive=False,
                        max_lines=1,
                    )
                    info_type = gr.Dropdown(choices=[], label=gettext("Select emotion"))
                with gr.Row():
                    add_voice_button = gr.Button(
                        gettext("Add new voice"), variant="primary"
                    )

                selected_chars_container = gr.Column(
                    elem_id="selected_chars_container", visible=False
                )

                with selected_chars_container:
                    gr.Markdown(gettext("### Selected characters"))
                    selected_chars_rows = []
                    for i in range(5):  # 假设最多选择5个角色
                        with gr.Row() as row:
                            name = gr.Textbox(
                                label=gettext("Name"), interactive=False, max_lines=1
                            )
                            emotion = gr.Textbox(
                                label=gettext("Emotion"), interactive=False, max_lines=1
                            )
                            delete_btn = gr.Button(gettext("Delete"), scale=0)
                        selected_chars_rows.append((name, emotion, delete_btn, row))

            with gr.Row():
                with gr.Column():
                    text = gr.Textbox(
                        label=gettext("Text to synthesize"),
                        value="",
                        lines=10,
                        max_lines=10,
                    )
                    inference_button = gr.Button(
                        gettext("🎉 Synthesize Voice 🎉"), variant="primary", size="lg"
                    )
                with gr.Column():
                    prompt_audio = gr.Audio(
                        label=gettext("Reference audio for synthesis"),
                        interactive=False,
                        type="numpy",
                    )
                    output = gr.Audio(
                        label=gettext("Output audio"), interactive=False, type="numpy"
                    )
                    cost_time = gr.Textbox(
                        label=gettext("Synthesis time"),
                        interactive=False,
                        show_label=False,
                        max_lines=1,
                    )
            try:
                inference_button.click(
                    fn=generate,
                    inputs=[current_character, selected_characters, text, lang],
                    outputs=[output, cost_time],
                )
            except gr.Error as e:
                gr.Error(e)
            except Exception as e:
                pass

        with gr.Tab(gettext("Create Voice")):
            with gr.Row():
                avatar = gr.Image(label=gettext("Avatar"), interactive=True, type="pil", image_mode="RGBA")
                with gr.Column():
                    with gr.Row():
                        name = gr.Textbox(
                            label=gettext("Name"), interactive=True, max_lines=1
                        )
                        emotion = gr.Textbox(
                            label=gettext("Emotion\n(Happy, Sad, Angry)"), interactive=True, max_lines=1
                        )
                        tags = gr.Textbox(
                            label=gettext("Tags\n(Genshin, Cute, Girl, Boy, etc.)"), interactive=True, max_lines=1
                        )
                        gender = gr.Dropdown(
                            label=gettext("Gender"),
                            choices=[
                                (gettext("Male"), "male"),
                                (gettext("Female"), "female"),
                                (gettext("Non-Binary"), "non-binary"),
                            ],
                            interactive=True,
                        )
                    audio_data = gr.Audio(label=gettext("Prompt Audio(min 3.2s, max 8s)"), interactive=True)
                    create_button = gr.Button(gettext("Create Voice"), variant="primary")
            
        gr.Markdown(gettext(terms))
        # -------------- 绑定事件 --------------

        lang.change(
            fn=update_all_characters,
            inputs=[lang, category],
            outputs=[all_characters_state, characters, gallery, category, kind],
        )

        add_voice_button.click(
            fn=add_new_voice,
            inputs=[
                current_character,
                selected_characters,
                kind,
                lang,
                all_characters_state,
            ],
            outputs=[
                current_character,
                character_name,
                info_type,
                selected_characters,
                characters,
                gallery,
                selected_chars_container,
                all_characters_state,
            ],
        ).then(
            fn=update_selected_chars_display,
            inputs=[selected_characters],
            outputs=[item for row in selected_chars_rows for item in row],
        )

        gallery.select(
            fn=on_select,
            inputs=[characters, selected_characters, all_characters_state],
            outputs=[character_name, info_type, current_character, selected_characters],
        ).then(
            fn=update_prompt_audio, inputs=[current_character], outputs=[prompt_audio]
        )

        info_type.change(
            fn=update_character_info,
            inputs=[character_name, info_type, current_character, all_characters_state],
            outputs=[current_character, all_characters_state],
        ).then(
            fn=update_prompt_audio, inputs=[current_character], outputs=[prompt_audio]
        )

        for i, (_, _, delete_btn, _) in enumerate(selected_chars_rows):
            delete_btn.click(
                fn=remove_character,
                inputs=[gr.Number(value=i, visible=False), selected_characters],
                outputs=[selected_characters, selected_chars_container],
            ).then(
                fn=update_selected_chars_display,
                inputs=[selected_characters],
                outputs=[item for row in selected_chars_rows for item in row],
            )

        kind.change(
            fn=update_gallery,
            inputs=[kind, query, all_characters_state],
            outputs=[characters, gallery, all_characters_state],
        )

        query.change(
            fn=update_gallery,
            inputs=[kind, query, all_characters_state],
            outputs=[characters, gallery, all_characters_state],
        )

        create_button.click(
            fn=create_voice,
            inputs=[avatar, name, emotion, tags, gender, audio_data, lang],
            outputs=[avatar, name, emotion, tags, gender, audio_data],
        )


if __name__ == "__main__":
    demo.queue(default_concurrency_limit=None).launch(
        show_api=False
    )