File size: 7,635 Bytes
7723c7e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1a78dc2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7723c7e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1a78dc2
7723c7e
 
1a78dc2
 
7723c7e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6092cc1
7723c7e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b437d8e
7723c7e
 
 
 
 
 
 
6092cc1
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
# coding=utf-8
import time
import os
import gradio as gr
import utils
import argparse
import commons
from models import SynthesizerTrn
from text import text_to_sequence
import torch
from torch import no_grad, LongTensor
import webbrowser
import logging
logging.getLogger('numba').setLevel(logging.WARNING)
limitation = os.getenv("SYSTEM") == "spaces"  # limit text and audio length in huggingface spaces

def get_text(text, hps):
    text_norm, clean_text = text_to_sequence(text, hps.symbols, 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 vits(text, language, speaker_id, noise_scale, noise_scale_w, length_scale):
    print(f'________________{speaker_id}')
    parser = argparse.ArgumentParser()
    parser.add_argument('--device', type=str, default='cpu')
    parser.add_argument('--api', action="store_true", default=True)
    parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
    parser.add_argument("--colab", action="store_true", default=False, help="share gradio app")
    args = parser.parse_args()
    device = torch.device(args.device)
    
    hps_ms = utils.get_hparams_from_file(r'./model/config.json')
    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,
        **hps_ms.model)
    _ = net_g_ms.eval().to(device)
    speakers = hps_ms.speakers
    model, optimizer, learning_rate, epochs = utils.load_checkpoint(r'./model/G_953000.pth', net_g_ms, None)
    
    
    start = time.perf_counter()
    if not len(text):
        return "输入文本不能为空!", None, None
    text = text.replace('\n', ' ').replace('\r', '').replace(" ", "")
    if len(text) > 100 and limitation:
        return f"输入文字过长!{len(text)}>100", None, None
    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)
    with no_grad():
        x_tst = stn_tst.unsqueeze(0).to(device)
        x_tst_lengths = LongTensor([stn_tst.size(0)]).to(device)
        speaker_id = LongTensor([speaker_id]).to(device)
        audio = net_g_ms.infer(x_tst, x_tst_lengths, sid=speaker_id, noise_scale=noise_scale, noise_scale_w=noise_scale_w,
                               length_scale=length_scale)[0][0, 0].data.cpu().float().numpy()

    return "生成成功!",(22050, audio), f"生成耗时 {round(time.perf_counter()-start, 2)} s"

def search_speaker(search_value):
    hps_ms = utils.get_hparams_from_file(r'./model/config.json')
    speakers = hps_ms.speakers
    for s in speakers:
        if search_value == s:
            return s
    for s in speakers:
        if search_value in s:
            return s

def change_lang(language):
    if language == 0:
        return 0.6, 0.668, 1.2
    else:
        return 0.6, 0.668, 1.1

download_audio_js = """
() =>{{
    let root = document.querySelector("body > gradio-app");
    if (root.shadowRoot != null)
        root = root.shadowRoot;
    let audio = root.querySelector("#tts-audio").querySelector("audio");
    let text = root.querySelector("#input-text").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=True)
    parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
    parser.add_argument("--colab", action="store_true", default=False, help="share gradio app")
    args = parser.parse_args()
    device = torch.device(args.device)
    
    hps_ms = utils.get_hparams_from_file(r'./model/config.json')
    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,
        **hps_ms.model)
    _ = net_g_ms.eval().to(device)
    speakers = hps_ms.speakers
    model, optimizer, learning_rate, epochs = utils.load_checkpoint(r'./model/G_953000.pth', net_g_ms, None)
    
    with gr.Blocks() as app:
        gr.Markdown(
            "# <center> VITS语音在线合成demo\n"
            "# <center> 严禁将模型用于任何商业项目,否则后果自负\n"
            "<div align='center'>主要有赛马娘,原神中文,原神日语,崩坏3的音色</div>"
            '<div align="center"><a><font color="#dd0000">结果有随机性,语调可能很奇怪,可多次生成取最佳效果</font></a></div>'
            '<div align="center"><a><font color="#dd0000">标点符号会影响生成的结果</font></a></div>'
        )

        with gr.Tabs():
            with gr.TabItem("vits"):
                with gr.Row():
                    with gr.Column():
                        input_text = gr.Textbox(label="Text (100 words limitation) " if limitation else "Text", lines=5, value="今天晚上吃啥好呢。", elem_id=f"input-text")
                        lang = gr.Dropdown(label="Language", 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")
                        btn = gr.Button(value="Submit")
                        with gr.Row():
                            search = gr.Textbox(label="Search Speaker", lines=1)
                            btn2 = gr.Button(value="Search")
                        sid = gr.Dropdown(label="Speaker", choices=speakers, type="index", value=speakers[228])
                        with gr.Row():
                            ns = gr.Slider(label="noise_scale(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="noise_scale_w(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="length_scale(控制整体语速)", minimum=0.1, maximum=2.0, step=0.1, value=1.2, interactive=True)
                    with gr.Column():
                        o1 = gr.Textbox(label="Output Message")
                        o2 = gr.Audio(label="Output Audio", elem_id=f"tts-audio")
                        o3 = gr.Textbox(label="Extra Info")
                        download = gr.Button("Download Audio")
                    btn.click(vits, inputs=[input_text, lang, sid, ns, nsw, ls], outputs=[o1, o2, o3],api_name='voice')
                    download.click(None, [], [], _js=download_audio_js.format())
                    btn2.click(search_speaker, inputs=[search], outputs=[sid])
                    lang.change(change_lang, inputs=[lang], outputs=[ns, nsw, ls])
            with gr.TabItem("可用人物一览"):
                gr.Radio(label="Speaker", choices=speakers, interactive=False, type="index")
    if args.colab:
        webbrowser.open("http://127.0.0.1:7860")
    app.queue(concurrency_count=3, api_open=True).launch(show_api=True)