File size: 6,518 Bytes
e921a23
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3a27cd1
e921a23
3a27cd1
e9f7ddf
3a27cd1
e921a23
3a27cd1
 
 
e921a23
 
 
 
3a27cd1
e921a23
3a27cd1
 
 
e921a23
 
 
 
 
 
 
 
 
 
 
 
 
ed5fdf5
b3f62b8
e921a23
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b0200dc
e921a23
 
 
 
 
 
 
 
 
 
566af11
e921a23
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e9f7ddf
e921a23
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
faf546b
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
import json
import os
import re

import librosa
import numpy as np
import torch
from torch import no_grad, LongTensor
import commons
import utils
import gradio as gr
from models import SynthesizerTrn
from text import text_to_sequence, _clean_text
from mel_processing import spectrogram_torch

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


def get_text(text, hps, is_phoneme):
    text_norm = text_to_sequence(text, hps.symbols, [] if is_phoneme 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_tts_fn(model, hps, speaker_ids):
    def tts_fn(text, speaker, speed, is_phoneme):
        if limitation:
            text_len = len(text)
            max_len = 700
            if is_phoneme:
                max_len *= 3
            else:
                if len(hps.data.text_cleaners) > 0 and hps.data.text_cleaners[0] == "zh_ja_mixture_cleaners":
                    text_len = len(re.sub("(\[ZH\]|\[JA\])", "", text))
            if text_len > max_len:
                return "Error: Text is too long", None

        speaker_id = speaker_ids[speaker]
        stn_tst = get_text(text, hps, is_phoneme)
        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 "Success", (hps.data.sampling_rate, audio)

    return tts_fn


def create_to_phoneme_fn(hps):
    def to_phoneme_fn(text):
        return _clean_text(text, hps.data.text_cleaners) if text != "" else ""

    return to_phoneme_fn
    
    
css = """
        #advanced-btn {
            color: white;
            border-color: black;
            background: black;
            font-size: .7rem !important;
            line-height: 19px;
            margin-top: 24px;
            margin-bottom: 12px;
            padding: 2px 8px;
            border-radius: 14px !important;
        }
        #advanced-options {
            display: none;
            margin-bottom: 20px;
        }
"""

if __name__ == '__main__':
    models_tts = []
    name = 'RaidenTTS'
    lang = '한국어 (Korean)'
    example = '내가 누군가를 좋아한다는 사실이 그 사람에게는 상처가 될 수 있잖아요.'
    config_path = f"saved_model/config.json"
    model_path = f"saved_model/model.pth"
    cover_path = f"saved_model/cover.png"
    hps = utils.get_hparams_from_file(config_path)
    model = SynthesizerTrn(
        len(hps.symbols),
        hps.data.filter_length // 2 + 1,
        hps.train.segment_size // hps.data.hop_length,
        n_speakers=hps.data.n_speakers,
        **hps.model)
    utils.load_checkpoint(model_path, model, None)
    model.eval()
    speaker_ids = [0]
    speakers = [name]

    t = 'vits'
    models_tts.append((name, cover_path, speakers, lang, example,
                        hps.symbols, create_tts_fn(model, hps, speaker_ids),
                        create_to_phoneme_fn(hps)))
                               
    app = gr.Blocks(css=css)

    with app:
        gr.Markdown("# Genshin Impact RaidenTTS Using Vits Model\n"
                    "![visitor badge](https://visitor-badge.glitch.me/badge?page_id=ORI-Muchim.RaidenTTS)\n\n")
        
        for i, (name, cover_path, speakers, lang, example, symbols, tts_fn,
                to_phoneme_fn) in enumerate(models_tts):

            with gr.Column():
                gr.Markdown(f"## {name}\n\n"
                            f"![cover](file/{cover_path})\n\n"
                            f"lang: {lang}")
                tts_input1 = gr.TextArea(label="Text (700 words limitation)", value=example,
                                            elem_id=f"tts-input{i}")
                tts_input2 = gr.Dropdown(label="Speaker", choices=speakers,
                                            type="index", value=speakers[0])
                tts_input3 = gr.Slider(label="Speed", value=1, minimum=0.1, maximum=2, step=0.1)
                with gr.Accordion(label="Advanced Options", open=False):
                    phoneme_input = gr.Checkbox(value=False, label="Phoneme input")
                    to_phoneme_btn = gr.Button("Covert text to phoneme")
                    phoneme_list = gr.Dataset(label="Phoneme list", components=[tts_input1],
                                                samples=[[x] for x in symbols],
                                                elem_id=f"phoneme-list{i}")
                    phoneme_list_json = gr.Json(value=symbols, visible=False)
                tts_submit = gr.Button("Generate", variant="primary")
                tts_output1 = gr.Textbox(label="Output Message")
                tts_output2 = gr.Audio(label="Output Audio")
                tts_submit.click(tts_fn, [tts_input1, tts_input2, tts_input3, phoneme_input],
                                    [tts_output1, tts_output2])
                to_phoneme_btn.click(to_phoneme_fn, [tts_input1], [tts_input1])
                phoneme_list.click(None, [phoneme_list, phoneme_list_json], [],
                                    _js=f"""
                (i,phonemes) => {{
                    let root = document.querySelector("body > gradio-app");
                    if (root.shadowRoot != null)
                        root = root.shadowRoot;
                    let text_input = root.querySelector("#tts-input{i}").querySelector("textarea");
                    let startPos = text_input.selectionStart;
                    let endPos = text_input.selectionEnd;
                    let oldTxt = text_input.value;
                    let result = oldTxt.substring(0, startPos) + phonemes[i] + oldTxt.substring(endPos);
                    text_input.value = result;
                    let x = window.scrollX, y = window.scrollY;
                    text_input.focus();
                    text_input.selectionStart = startPos + phonemes[i].length;
                    text_input.selectionEnd = startPos + phonemes[i].length;
                    text_input.blur();
                    window.scrollTo(x, y);
                    return [];
                }}""")

    app.queue(concurrency_count=3).launch(show_api=False)