File size: 15,337 Bytes
b875bd7
 
 
3fd0c0d
b875bd7
 
 
3fd0c0d
b875bd7
 
8e15dd6
b875bd7
b54579f
b875bd7
 
3fd0c0d
 
b875bd7
 
b54579f
 
 
 
 
 
 
 
 
 
 
8e15dd6
 
b875bd7
 
 
 
 
 
8e15dd6
b875bd7
3fd0c0d
 
 
a9e3d65
 
3fd0c0d
 
8e15dd6
 
 
 
 
 
 
 
b875bd7
3fd0c0d
 
 
b875bd7
3fd0c0d
b875bd7
 
 
 
8e15dd6
 
 
 
 
 
 
 
 
 
 
b875bd7
 
8e15dd6
 
 
b875bd7
8e15dd6
b875bd7
 
 
 
 
 
d95ac39
 
b875bd7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3fd0c0d
 
d595ee0
3fd0c0d
 
 
 
b875bd7
 
 
 
51ff334
 
3fd0c0d
 
 
 
 
b54579f
 
b875bd7
 
 
 
75b5bd6
b875bd7
 
3fd0c0d
8e15dd6
b875bd7
 
 
1500087
8e15dd6
b875bd7
b3bac56
 
b54579f
4133e57
b875bd7
 
 
 
8e15dd6
b875bd7
 
 
 
 
f4da48c
b875bd7
 
 
 
a9e3d65
b875bd7
b54579f
8e15dd6
 
 
 
 
 
 
 
b875bd7
 
 
a974a24
b875bd7
 
d95ac39
b875bd7
0c3c2eb
8e15dd6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b875bd7
8e15dd6
b875bd7
 
 
 
 
f4da48c
b875bd7
 
 
 
a9e3d65
b875bd7
b54579f
8e15dd6
 
 
 
 
 
 
 
b875bd7
 
 
a974a24
b875bd7
 
d95ac39
b875bd7
0c3c2eb
d95ac39
b875bd7
8e15dd6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d595ee0
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
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
# coding=utf-8
import os
import re
import argparse
import utils
import commons
import json
import torch
import gradio as gr
from models import SynthesizerTrn
from text import text_to_sequence, _clean_text
from torch import no_grad, LongTensor
import gradio.processing_utils as gr_processing_utils
import logging
logging.getLogger('numba').setLevel(logging.WARNING)
limitation = os.getenv("SYSTEM") == "spaces"  # limit text and audio length in huggingface spaces

hps_ms = utils.get_hparams_from_file(r'config/config.json')

audio_postprocess_ori = gr.Audio.postprocess

def audio_postprocess(self, y):
    data = audio_postprocess_ori(self, y)
    if data is None:
        return None
    return gr_processing_utils.encode_url_or_file_to_base64(data["name"])


gr.Audio.postprocess = audio_postprocess

def get_text(text, hps, is_symbol):
    text_norm, clean_text = 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, clean_text

def create_tts_fn(net_g_ms, speaker_id):
    def tts_fn(text, language, noise_scale, noise_scale_w, length_scale, is_symbol):
        text = text.replace('\n', ' ').replace('\r', '').replace(" ", "")
        if limitation:
            text_len = len(re.sub("\[([A-Z]{2})\]", "", text))
            max_len = 100
            if is_symbol:
                max_len *= 3
            if text_len > max_len:
                return "Error: Text is too long", None
        if not is_symbol:
            if language == 0:
                text = f"[ZH]{text}[ZH]"
            elif language == 1:
                text = f"[JA]{text}[JA]"
            else:
                text = f"{text}"
        stn_tst, clean_text = get_text(text, hps_ms, is_symbol)
        with no_grad():
            x_tst = stn_tst.unsqueeze(0).to(device)
            x_tst_lengths = LongTensor([stn_tst.size(0)]).to(device)
            sid = LongTensor([speaker_id]).to(device)
            audio = net_g_ms.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=noise_scale, noise_scale_w=noise_scale_w,
                                   length_scale=length_scale)[0][0, 0].data.cpu().float().numpy()

        return "Success", (22050, audio)
    return tts_fn

def create_to_symbol_fn(hps):
    def to_symbol_fn(is_symbol_input, input_text, temp_text, temp_lang):
        if temp_lang == 'Chinese':
            clean_text = f'[ZH]{input_text}[ZH]'
        elif temp_lang == "Japanese":
            clean_text = f'[JA]{input_text}[JA]'
        else:
            clean_text = input_text
        return (_clean_text(clean_text, hps.data.text_cleaners), input_text) if is_symbol_input else (temp_text, temp_text)

    return to_symbol_fn
def change_lang(language):
    if language == 0:
        return 0.6, 0.668, 1.2, "Chinese"
    elif language == 1:
        return 0.6, 0.668, 1, "Japanese"
    else:
        return 0.6, 0.668, 1, "Mix"

download_audio_js = """
() =>{{
    let root = document.querySelector("body > gradio-app");
    if (root.shadowRoot != null)
        root = root.shadowRoot;
    let audio = root.querySelector("#tts-audio-{audio_id}").querySelector("audio");
    let text = root.querySelector("#input-text-{audio_id}").querySelector("textarea");
    if (audio == undefined)
        return;
    text = text.value;
    if (text == undefined)
        text = Math.floor(Math.random()*100000000);
    audio = audio.src;
    let oA = document.createElement("a");
    oA.download = text.substr(0, 20)+'.wav';
    oA.href = audio;
    document.body.appendChild(oA);
    oA.click();
    oA.remove();
}}
"""

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--device', type=str, default='cpu')
    parser.add_argument('--api', action="store_true", default=False)
    parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
    args = parser.parse_args()
    device = torch.device(args.device)
    
    models = []
    with open("pretrained_models/info.json", "r", encoding="utf-8") as f:
        models_info = json.load(f)
    for i, info in models_info.items():
        if not info['enable']:
            continue
        sid = info['sid']
        name_en = info['name_en']
        name_zh = info['name_zh']
        title = info['title']
        cover = f"pretrained_models/{i}/{info['cover']}"
        example = info['example']
        language = info['language']
        net_g_ms = SynthesizerTrn(
            len(hps_ms.symbols),
            hps_ms.data.filter_length // 2 + 1,
            hps_ms.train.segment_size // hps_ms.data.hop_length,
            n_speakers=hps_ms.data.n_speakers if info['type'] == "multi" else 0,
            **hps_ms.model)
        utils.load_checkpoint(f'pretrained_models/{i}/{i}.pth', net_g_ms, None)
        _ = net_g_ms.eval().to(device)
        models.append((sid, name_en, name_zh, title, cover, example, language, net_g_ms, create_tts_fn(net_g_ms, sid), create_to_symbol_fn(hps_ms)))
    with gr.Blocks() as app:
        gr.Markdown(
            "# <center> vits-models\n"
            "## <center> Please do not generate content that could infringe upon the rights or cause harm to individuals or organizations.\n"
            "## <center> ·请不要生成会对个人以及组织造成侵害的内容\n"
            "![visitor badge](https://visitor-badge.glitch.me/badge?page_id=sayashi.vits-models)\n\n"
            "[Open In Colab]"
            "(https://colab.research.google.com/drive/10QOk9NPgoKZUXkIhhuVaZ7SYra1MPMKH?usp=share_link)"
            " without queue and length limitation.(无需等待队列,并且没有长度限制)\n\n"
            "[Finetune your own model](https://github.com/SayaSS/vits-finetuning)"
        )

        with gr.Tabs():
            with gr.TabItem("EN"):
                for (sid, name_en, name_zh, title, cover, example, language, net_g_ms, tts_fn, to_symbol_fn) in models:
                    with gr.TabItem(name_en):
                        with gr.Row():
                            gr.Markdown(
                                '<div align="center">'
                                f'<a><strong>{title}</strong></a>'
                                f'<img style="width:auto;height:300px;" src="file/{cover}">' if cover else ""
                                '</div>'
                            )
                        with gr.Row():
                            with gr.Column():
                                input_text = gr.Textbox(label="Text (100 words limitation)" if limitation else "Text", lines=5, value=example, elem_id=f"input-text-en-{name_en.replace(' ','')}")
                                lang = gr.Dropdown(label="Language", choices=["Chinese", "Japanese", "Mix(wrap the Chinese text with [ZH][ZH], wrap the Japanese text with [JA][JA])"],
                                            type="index", value=language)
                                temp_lang = gr.Variable(value=language)
                                with gr.Accordion(label="Advanced Options", open=False):
                                    temp_text_var = gr.Variable()
                                    symbol_input = gr.Checkbox(value=False, label="Symbol input")
                                    symbol_list = gr.Dataset(label="Symbol list", components=[input_text],
                                                             samples=[[x] for x in hps_ms.symbols])
                                    symbol_list_json = gr.Json(value=hps_ms.symbols, visible=False)
                                btn = gr.Button(value="Generate", variant="primary")
                                with gr.Row():
                                    ns = gr.Slider(label="noise_scale", minimum=0.1, maximum=1.0, step=0.1, value=0.6, interactive=True)
                                    nsw = gr.Slider(label="noise_scale_w", minimum=0.1, maximum=1.0, step=0.1, value=0.668, interactive=True)
                                    ls = gr.Slider(label="length_scale", minimum=0.1, maximum=2.0, step=0.1, value=1.2 if language=="Chinese" else 1, interactive=True)
                            with gr.Column():
                                o1 = gr.Textbox(label="Output Message")
                                o2 = gr.Audio(label="Output Audio", elem_id=f"tts-audio-en-{name_en.replace(' ','')}")
                                download = gr.Button("Download Audio")
                            btn.click(tts_fn, inputs=[input_text, lang,  ns, nsw, ls, symbol_input], outputs=[o1, o2], api_name=f"tts-{name_en}")
                            download.click(None, [], [], _js=download_audio_js.format(audio_id=f"en-{name_en.replace(' ', '')}"))
                            lang.change(change_lang, inputs=[lang], outputs=[ns, nsw, ls, temp_lang])
                            symbol_input.change(
                                to_symbol_fn,
                                [symbol_input, input_text, temp_text_var, temp_lang],
                                [input_text, temp_text_var]
                            )
                            symbol_list.click(None, [symbol_list, symbol_list_json], [input_text],
                                              _js=f"""
                            (i,symbols) => {{
                                let root = document.querySelector("body > gradio-app");
                                if (root.shadowRoot != null)
                                    root = root.shadowRoot;
                                let text_input = root.querySelector("#input-text-en-{name_en.replace(' ', '')}").querySelector("textarea");
                                let startPos = text_input.selectionStart;
                                let endPos = text_input.selectionEnd;
                                let oldTxt = text_input.value;
                                let result = oldTxt.substring(0, startPos) + symbols[i] + oldTxt.substring(endPos);
                                text_input.value = result;
                                let x = window.scrollX, y = window.scrollY;
                                text_input.focus();
                                text_input.selectionStart = startPos + symbols[i].length;
                                text_input.selectionEnd = startPos + symbols[i].length;
                                text_input.blur();
                                window.scrollTo(x, y);
                                return text_input.value;
                            }}""")
            with gr.TabItem("中文"):
                for (sid, name_en, name_zh, title, cover, example, language,  net_g_ms, tts_fn, to_symbol_fn) in models:
                    with gr.TabItem(name_zh):
                        with gr.Row():
                            gr.Markdown(
                                '<div align="center">'
                                f'<a><strong>{title}</strong></a>'
                                f'<img style="width:auto;height:300px;" src="file/{cover}">' if cover else ""
                                '</div>'
                            )
                        with gr.Row():
                            with gr.Column():
                                input_text = gr.Textbox(label="文本 (100字上限)" if limitation else "文本", lines=5, value=example, elem_id=f"input-text-zh-{name_zh}")
                                lang = gr.Dropdown(label="语言", choices=["中文", "日语", "中日混合(中文用[ZH][ZH]包裹起来,日文用[JA][JA]包裹起来)"],
                                            type="index", value="中文"if language == "Chinese" else "日语")
                                temp_lang = gr.Variable(value=language)
                                with gr.Accordion(label="高级选项", open=False):
                                    temp_text_var = gr.Variable()
                                    symbol_input = gr.Checkbox(value=False, label="符号输入")
                                    symbol_list = gr.Dataset(label="符号列表", components=[input_text],
                                                             samples=[[x] for x in hps_ms.symbols])
                                    symbol_list_json = gr.Json(value=hps_ms.symbols, visible=False)
                                btn = gr.Button(value="生成", variant="primary")
                                with gr.Row():
                                    ns = gr.Slider(label="控制感情变化程度", minimum=0.1, maximum=1.0, step=0.1, value=0.6, interactive=True)
                                    nsw = gr.Slider(label="控制音素发音长度", minimum=0.1, maximum=1.0, step=0.1, value=0.668, interactive=True)
                                    ls = gr.Slider(label="控制整体语速", minimum=0.1, maximum=2.0, step=0.1, value=1.2 if language=="Chinese" else 1, interactive=True)
                            with gr.Column():
                                o1 = gr.Textbox(label="输出信息")
                                o2 = gr.Audio(label="输出音频", elem_id=f"tts-audio-zh-{name_zh}")
                                download = gr.Button("下载音频")
                            btn.click(tts_fn, inputs=[input_text, lang,  ns, nsw, ls, symbol_input], outputs=[o1, o2])
                            download.click(None, [], [], _js=download_audio_js.format(audio_id=f"zh-{name_zh}"))
                            lang.change(change_lang, inputs=[lang], outputs=[ns, nsw, ls])
                            symbol_input.change(
                                to_symbol_fn,
                                [symbol_input, input_text, temp_text_var, temp_lang],
                                [input_text, temp_text_var]
                            )
                            symbol_list.click(None, [symbol_list, symbol_list_json], [input_text],
                                              _js=f"""
                            (i,symbols) => {{
                                let root = document.querySelector("body > gradio-app");
                                if (root.shadowRoot != null)
                                    root = root.shadowRoot;
                                let text_input = root.querySelector("#input-text-zh-{name_zh}").querySelector("textarea");
                                let startPos = text_input.selectionStart;
                                let endPos = text_input.selectionEnd;
                                let oldTxt = text_input.value;
                                let result = oldTxt.substring(0, startPos) + symbols[i] + oldTxt.substring(endPos);
                                text_input.value = result;
                                let x = window.scrollX, y = window.scrollY;
                                text_input.focus();
                                text_input.selectionStart = startPos + symbols[i].length;
                                text_input.selectionEnd = startPos + symbols[i].length;
                                text_input.blur();
                                window.scrollTo(x, y);
                                return text_input.value;
                            }}""")
    app.queue(concurrency_count=1, api_open=args.api).launch(share=args.share)