import argparse
import json
import os
import re
import tempfile
import logging

logging.getLogger('numba').setLevel(logging.WARNING)
import librosa
import numpy as np
import torch
from torch import no_grad, LongTensor
import commons
import utils
import gradio as gr
import ONNXVITS_infer
import models
from text import text_to_sequence, _clean_text
from text.symbols import symbols
from mel_processing import spectrogram_torch
import psutil
from datetime import datetime

language_marks = {
    "Japanese": "",
    "日本語": "[JA]",
    "简体中文": "[ZH]",
    "English": "[EN]",
    "Mix": "",
}

limitation = os.getenv("SYSTEM") == "spaces"  # limit text and audio length in huggingface spaces


def create_tts_fn(model, hps, speaker_ids):
    def tts_fn(text, speaker, language, speed, is_symbol):
        if limitation:
            text_len = len(re.sub("\[([A-Z]{2})\]", "", text))
            max_len = 500  #文本长度限制
            if is_symbol:
                max_len *= 3
            if text_len > max_len:
                return "错误：文本太长", None
        if language is not None:
            text = language_marks[language] + text + language_marks[language]
        speaker_id = speaker_ids[speaker]
        stn_tst = get_text(text, hps, is_symbol)
        with no_grad():
            x_tst = stn_tst.unsqueeze(0)
            x_tst_lengths = LongTensor([stn_tst.size(0)])
            sid = LongTensor([speaker_id])
            audio = model.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=.667, noise_scale_w=0.8,
                                length_scale=1.0 / speed)[0][0, 0].data.cpu().float().numpy()
        del stn_tst, x_tst, x_tst_lengths, sid
        return "成功", (hps.data.sampling_rate, audio)

    return tts_fn


def create_vc_fn(model, hps, speaker_ids):
    def vc_fn(original_speaker, target_speaker, input_audio):
        if input_audio is None:
            return "您需要上传音频", None
        sampling_rate, audio = input_audio
        duration = audio.shape[0] / sampling_rate
        if limitation and duration > 30: #限制音频长度
            return "错误：音频太长", None
        original_speaker_id = speaker_ids[original_speaker]
        target_speaker_id = speaker_ids[target_speaker]

        audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
        if len(audio.shape) > 1:
            audio = librosa.to_mono(audio.transpose(1, 0))
        if sampling_rate != hps.data.sampling_rate:
            audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=hps.data.sampling_rate)
        with no_grad():
            y = torch.FloatTensor(audio)
            y = y.unsqueeze(0)
            spec = spectrogram_torch(y, hps.data.filter_length,
                                     hps.data.sampling_rate, hps.data.hop_length, hps.data.win_length,
                                     center=False)
            spec_lengths = LongTensor([spec.size(-1)])
            sid_src = LongTensor([original_speaker_id])
            sid_tgt = LongTensor([target_speaker_id])
            audio = model.voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt)[0][
                0, 0].data.cpu().float().numpy()
        del y, spec, spec_lengths, sid_src, sid_tgt
        return "成功", (hps.data.sampling_rate, audio)

    return vc_fn


def get_text(text, hps, is_symbol):
    text = text.replace('\n', '[BR]')
    text_norm = text_to_sequence(text, hps.symbols, [] if is_symbol else hps.data.text_cleaners)
    if hps.data.add_blank:
        text_norm = commons.intersperse(text_norm, 0)
    text_norm = LongTensor(text_norm)
    return text_norm


def create_to_symbol_fn(hps):
    def to_symbol_fn(is_symbol_input, input_text, temp_text):
        return (_clean_text(input_text, hps.data.text_cleaners), input_text) if is_symbol_input \
            else (temp_text, temp_text)

    return to_symbol_fn


models_tts = []
models_vc = []
models_info = [
    {
        "title": "三种语言",
        "languages": ['日本語', '简体中文', 'English', 'Mix'],
        "description": """
    这个模型在赛马娘，原神，魔女的夜宴以及VCTK数据集上混合训练以学习多种语言。
    所有角色均可说中日英三语。\n\n
    若需要在同一个句子中混合多种语言，使用相应的语言标记包裹句子。
    （日语用[JA], 中文用[ZH], 英文用[EN]），参考Examples中的示例。
    """,
        "model_path": "./pretrained_models/G_trilingual.pth",
        "config_path": "./configs/uma_trilingual.json",
        "examples": [['你好，训练员先生，很高兴见到你。', '草上飞 Grass Wonder (Umamusume Pretty Derby)', '简体中文', 1, False],
                     ['To be honest, I have no idea what to say as examples.', '派蒙 Paimon (Genshin Impact)', 'English',
                      1, False],
                     ['授業中に出しだら，学校生活終わるですわ。', '綾地 寧々 Ayachi Nene (Sanoba Witch)', '日本語', 1, False],
                     ['[JA]こんにちわ。[JA][ZH]你好！[ZH][EN]Hello![EN]', '綾地 寧々 Ayachi Nene (Sanoba Witch)', 'Mix', 1, False]],
        "onnx_dir": "./ONNX_net/G_trilingual/"
    },
    {
        "title": "日语",
        "languages": ["Japanese"],
        "description": """
                       这个模型包含赛马娘的所有87名角色，只能合成日语。
                       """,
        "model_path": "./pretrained_models/G_jp.pth",
        "config_path": "./configs/uma87.json",
        "examples": [['お疲れ様です，トレーナーさん。', '无声铃鹿 Silence Suzuka (Umamusume Pretty Derby)', 'Japanese', 1, False],
                     ['張り切っていこう！', '北部玄驹 Kitasan Black (Umamusume Pretty Derby)', 'Japanese', 1, False],
                     ['何でこんなに慣れでんのよ，私のほが先に好きだっだのに。', '草上飞 Grass Wonder (Umamusume Pretty Derby)', 'Japanese', 1, False],
                     ['授業中に出しだら，学校生活終わるですわ。', '目白麦昆 Mejiro Mcqueen (Umamusume Pretty Derby)', 'Japanese', 1, False],
                     ['お帰りなさい，お兄様！', '米浴 Rice Shower (Umamusume Pretty Derby)', 'Japanese', 1, False],
                     ['私の処女をもらっでください！', '米浴 Rice Shower (Umamusume Pretty Derby)', 'Japanese', 1, False]],
        "onnx_dir": "./ONNX_net/G_jp/"
    },
]

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
    args = parser.parse_args()
    for info in models_info:
        name = info['title']
        lang = info['languages']
        examples = info['examples']
        config_path = info['config_path']
        model_path = info['model_path']
        description = info['description']
        onnx_dir = info["onnx_dir"]
        hps = utils.get_hparams_from_file(config_path)
        model = ONNXVITS_infer.SynthesizerTrn(
            len(hps.symbols),
            hps.data.filter_length // 2 + 1,
            hps.data.sampling_rate // hps.data.hop_length,
            n_speakers=hps.data.n_speakers,
            ONNX_dir=onnx_dir,
            **hps.model)
        utils.load_checkpoint(model_path, model, None)
        model.eval()
        speaker_ids = hps.speakers
        speakers = list(hps.speakers.keys())
        models_tts.append((name, description, speakers, lang, examples,
                           hps.symbols, create_tts_fn(model, hps, speaker_ids),
                           create_to_symbol_fn(hps)))
        models_vc.append((name, description, speakers, create_vc_fn(model, hps, speaker_ids)))
    app = gr.Blocks()
    with app:
        gr.Markdown("# 中文 & 英文 & 日语 动漫 TTS\n\n"
                    "## 此项目原作者一直没更新，我已更新到最新版的gradio，并修复已知问题，速度超快的好吧。\n\n包含一个纯日语TTS和一个中日英三语TTS模型，主要为二次元角色。"
                    )
        with gr.Tabs() as tabs:
            with gr.Tab("TTS"):
                with gr.Tabs() as model_tabs:
                    for i, (name, description, speakers, lang, example, symbols, tts_fn, to_symbol_fn) in enumerate(models_tts):
                        with gr.Tab(name):
                            gr.Markdown(description)
                            with gr.Row():
                                with gr.Column():
                                    textbox = gr.Textbox(
                                        label="Text",
                                        placeholder="在此处输入您的句子（最多 500 个单词）",
                                        value="こんにちわ。",
                                        elem_id=f"tts-input",
                                        lines=3
                                    )
                                    with gr.Accordion(label="Phoneme Input", open=False):
                                        temp_text_var = gr.State()
                                        symbol_input = gr.Checkbox(value=False, label="Symbol input")
                                        symbol_list = gr.Dataframe(
                                            headers=["Symbols"],
                                            value=[[x] for x in symbols],
                                            elem_id=f"symbol-list",
                                            interactive=False
                                        )
                                        symbol_list_json = gr.JSON(value=symbols, visible=False)
                                    
                                    symbol_input.change(
                                        fn=to_symbol_fn,
                                        inputs=[symbol_input, textbox, temp_text_var],
                                        outputs=[textbox, temp_text_var]
                                    )
                                    
                                    def update_text(evt):
                                        if evt is None:  # 处理空事件
                                            return ""
                                        return evt  # 直接返回事件值
                                    
                                    symbol_list.select(
                                        fn=update_text,
                                        outputs=textbox
                                    )
                                    
                                    gr.HTML(value="", elem_id="symbol-list-js")
                                    gr.HTML("""
                                        <script>
                                        document.addEventListener('DOMContentLoaded', function() {
                                            document.addEventListener('gradio_select', function(e) {
                                                if (e.target && e.target.id && e.target.id.includes('symbol-list')) {
                                                    const text_input = document.querySelector("#tts-input textarea");
                                                    if (text_input) {
                                                        const startPos = text_input.selectionStart;
                                                        const endPos = text_input.selectionEnd;
                                                        const oldTxt = text_input.value;
                                                        const symbol = e.detail.value;
                                                        const result = oldTxt.substring(0, startPos) + symbol + oldTxt.substring(endPos);
                                                        text_input.value = result;
                                                        text_input.focus();
                                                        text_input.selectionStart = startPos + symbol.length;
                                                        text_input.selectionEnd = startPos + symbol.length;
                                                    }
                                                }
                                            });
                                        });
                                        </script>
                                    """)
                                    
                                    char_dropdown = gr.Dropdown(
                                        choices=speakers,
                                        value=speakers[0],
                                        label='character'
                                    )
                                    language_dropdown = gr.Dropdown(
                                        choices=lang,
                                        value=lang[0],
                                        label='language'
                                    )
                                    duration_slider = gr.Slider(
                                        minimum=0.1,
                                        maximum=5,
                                        value=1,
                                        step=0.1,
                                        label='速度 Speed'
                                    )
                                
                                with gr.Column():
                                    text_output = gr.Textbox(label="Message")
                                    audio_output = gr.Audio(
                                        label="Output Audio",
                                        elem_id="tts-audio",
                                        type="numpy"
                                    )
                                    btn = gr.Button("生成!")
                                    btn.click(
                                        fn=tts_fn,
                                        inputs=[
                                            textbox,
                                            char_dropdown,
                                            language_dropdown,
                                            duration_slider,
                                            symbol_input
                                        ],
                                        outputs=[text_output, audio_output]
                                    )
                            
                            gr.Examples(
                                examples=example,
                                inputs=[
                                    textbox,
                                    char_dropdown,
                                    language_dropdown,
                                    duration_slider,
                                    symbol_input
                                ],
                                outputs=[text_output, audio_output],
                                fn=tts_fn
                            )
    
    app.queue(max_size=3).launch(show_api=True, share=args.share)