File size: 16,842 Bytes
32ed620
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0163c77
32ed620
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f177ee9
776b6ae
 
 
 
 
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
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
import argparse
import json
import os
import re
import tempfile
from pathlib import Path

import librosa
import numpy as np
import torch
from torch import no_grad, LongTensor
import commons
import utils
import gradio as gr
import gradio.utils as gr_utils
import gradio.processing_utils as gr_processing_utils
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

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 = 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_tts_fn(model, hps, speaker_ids):
    def tts_fn(text, speaker, speed, is_symbol):
        print("generating audio...")
        if limitation:
            text_len = len(re.sub("\[([A-Z]{2})\]", "", text))
            max_len = 150
            if is_symbol:
                max_len *= 3
            if text_len > max_len:
                return "Error: Text is too long", None

        speaker_id = speaker_ids[speaker]
        stn_tst = get_text(text, hps, 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 = 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_vc_fn(model, hps, speaker_ids):
    def vc_fn(original_speaker, target_speaker, input_audio):
        if input_audio is None:
            return "You need to upload an audio", None
        sampling_rate, audio = input_audio
        duration = audio.shape[0] / sampling_rate
        if limitation and duration > 30:
            return "Error: Audio is too long", 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).to(device)
            spec_lengths = LongTensor([spec.size(-1)]).to(device)
            sid_src = LongTensor([original_speaker_id]).to(device)
            sid_tgt = LongTensor([target_speaker_id]).to(device)
            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 "Success", (hps.data.sampling_rate, audio)

    return vc_fn


def create_soft_vc_fn(model, hps, speaker_ids):
    def soft_vc_fn(target_speaker, input_audio1, input_audio2):
        input_audio = input_audio1
        if input_audio is None:
            input_audio = input_audio2
        if input_audio is None:
            return "You need to upload an audio", None
        sampling_rate, audio = input_audio
        duration = audio.shape[0] / sampling_rate
        if limitation and duration > 30:
            return "Error: Audio is too long", None
        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 != 16000:
            audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
        with torch.inference_mode():
            units = hubert.units(torch.FloatTensor(audio).unsqueeze(0).unsqueeze(0).to(device))
        with no_grad():
            unit_lengths = LongTensor([units.size(1)]).to(device)
            sid = LongTensor([target_speaker_id]).to(device)
            audio = model.infer(units, unit_lengths, sid=sid, noise_scale=.667,
                                noise_scale_w=0.8)[0][0, 0].data.cpu().float().numpy()
        del units, unit_lengths, sid
        return "Success", (hps.data.sampling_rate, audio)

    return soft_vc_fn


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


download_audio_js = """
() =>{{
    let root = document.querySelector("body > gradio-app");
    if (root.shadowRoot != null)
        root = root.shadowRoot;
    let audio = root.querySelector("#{audio_id}").querySelector("audio");
    if (audio == undefined)
        return;
    audio = audio.src;
    let oA = document.createElement("a");
    oA.download = Math.floor(Math.random()*100000000)+'.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("--share", action="store_true", default=False, help="share gradio app")
    args = parser.parse_args()

    device = torch.device(args.device)
    models_tts = []
    models_vc = []
    models_soft_vc = []
    with open("saved_model/info.json", "r", encoding="utf-8") as f:
        models_info = json.load(f)
    for i, info in models_info.items():
        name = info["title"]
        author = info["author"]
        lang = info["lang"]
        example = info["example"]
        config_path = f"saved_model/{i}/config.json"
        model_path = f"saved_model/{i}/model.pth"
        cover = info["cover"]
        cover_path = f"saved_model/{i}/{cover}" if cover else None
        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().to(device)
        speaker_ids = [sid for sid, name in enumerate(hps.speakers) if name != "None"]
        speakers = [name for sid, name in enumerate(hps.speakers) if name != "None"]

        t = info["type"]
        if t == "vits":
            models_tts.append((name, author, cover_path, speakers, lang, example,
                               hps.symbols, create_tts_fn(model, hps, speaker_ids),
                               create_to_symbol_fn(hps)))
            models_vc.append((name, author, cover_path, speakers, create_vc_fn(model, hps, speaker_ids)))
        elif t == "soft-vits-vc":
            models_soft_vc.append((name, author, cover_path, speakers, create_soft_vc_fn(model, hps, speaker_ids)))

    hubert = torch.hub.load("bshall/hubert:main", "hubert_soft", trust_repo=True).to(device)

    app = gr.Blocks()

    with app:
        gr.Markdown("# Moe TTS And Voice Conversion Using VITS Model\n\n"
                    "![visitor badge](https://visitor-badge.glitch.me/badge?page_id=skytnt.moegoe)\n\n"
                    "[Open In Colab]"
                    "(https://colab.research.google.com/drive/14Pb8lpmwZL-JI5Ub6jpG4sz2-8KS0kbS?usp=sharing)"
                    " without queue and length limitation.\n\n"
                    "Feel free to [open discussion](https://huggingface.co/spaces/skytnt/moe-tts/discussions/new) "
                    "if you want to add your model to this app.")
        with gr.Tabs():
            with gr.TabItem("TTS"):
                with gr.Tabs():
                    for i, (name, author, cover_path, speakers, lang, example, symbols, tts_fn,
                            to_symbol_fn) in enumerate(models_tts):
                        with gr.TabItem(f"model{i}"):
                            with gr.Column():
                                cover_markdown = f"![cover](file/{cover_path})\n\n" if cover_path else ""
                                gr.Markdown(f"## {name}\n\n"
                                            f"{cover_markdown}"
                                            f"model author: {author}\n\n"
                                            f"language: {lang}")
                                tts_input1 = gr.TextArea(label="Text (150 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=0.5, maximum=2, step=0.1)
                                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=[tts_input1],
                                                             samples=[[x] for x in symbols],
                                                             elem_id=f"symbol-list{i}")
                                    symbol_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", elem_id=f"tts-audio{i}")
                                download = gr.Button("Download Audio")
                                download.click(None, [], [], _js=download_audio_js.format(audio_id=f"tts-audio{i}"))

                                tts_submit.click(tts_fn, [tts_input1, tts_input2, tts_input3, symbol_input],
                                                 [tts_output1, tts_output2])
                                symbol_input.change(to_symbol_fn,
                                                    [symbol_input, tts_input1, temp_text_var],
                                                    [tts_input1, temp_text_var])
                                symbol_list.click(None, [symbol_list, symbol_list_json], [],
                                                  _js=f"""
                                (i,symbols) => {{
                                    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) + 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 [];
                                }}""")

            with gr.TabItem("Voice Conversion"):
                with gr.Tabs():
                    for i, (name, author, cover_path, speakers, vc_fn) in enumerate(models_vc):
                        with gr.TabItem(f"model{i}"):
                            cover_markdown = f"![cover](file/{cover_path})\n\n" if cover_path else ""
                            gr.Markdown(f"## {name}\n\n"
                                        f"{cover_markdown}"
                                        f"model author: {author}")
                            vc_input1 = gr.Dropdown(label="Original Speaker", choices=speakers, type="index",
                                                    value=speakers[0])
                            vc_input2 = gr.Dropdown(label="Target Speaker", choices=speakers, type="index",
                                                    value=speakers[min(len(speakers) - 1, 1)])
                            vc_input3 = gr.Audio(label="Input Audio (30s limitation)")
                            vc_submit = gr.Button("Convert", variant="primary")
                            vc_output1 = gr.Textbox(label="Output Message")
                            vc_output2 = gr.Audio(label="Output Audio", elem_id=f"vc-audio{i}")
                            download = gr.Button("Download Audio")
                            download.click(None, [], [], _js=download_audio_js.format(audio_id=f"vc-audio{i}"))
                            vc_submit.click(vc_fn, [vc_input1, vc_input2, vc_input3], [vc_output1, vc_output2])
            with gr.TabItem("Soft Voice Conversion"):
                with gr.Tabs():
                    for i, (name, author, cover_path, speakers, soft_vc_fn) in enumerate(models_soft_vc):
                        with gr.TabItem(f"model{i}"):
                            cover_markdown = f"![cover](file/{cover_path})\n\n" if cover_path else ""
                            gr.Markdown(f"## {name}\n\n"
                                        f"{cover_markdown}"
                                        f"model author: {author}")
                            vc_input1 = gr.Dropdown(label="Target Speaker", choices=speakers, type="index",
                                                    value=speakers[0])
                            source_tabs = gr.Tabs()
                            with source_tabs:
                                with gr.TabItem("microphone"):
                                    vc_input2 = gr.Audio(label="Input Audio (30s limitation)", source="microphone")
                                with gr.TabItem("upload"):
                                    vc_input3 = gr.Audio(label="Input Audio (30s limitation)", source="upload")
                            vc_submit = gr.Button("Convert", variant="primary")
                            vc_output1 = gr.Textbox(label="Output Message")
                            vc_output2 = gr.Audio(label="Output Audio", elem_id=f"svc-audio{i}")
                            download = gr.Button("Download Audio")
                            download.click(None, [], [], _js=download_audio_js.format(audio_id=f"svc-audio{i}"))
                            # clear inputs
                            source_tabs.set_event_trigger("change", None, [], [vc_input2, vc_input3],
                                                          js="()=>[null,null]")
                            vc_submit.click(soft_vc_fn, [vc_input1, vc_input2, vc_input3],
                                            [vc_output1, vc_output2])
        gr.Markdown(
            "unofficial demo for \n\n"
            "- [https://github.com/CjangCjengh/MoeGoe](https://github.com/CjangCjengh/MoeGoe)\n"
            "- [https://github.com/Francis-Komizu/VITS](https://github.com/Francis-Komizu/VITS)\n"
            "- [https://github.com/luoyily/MoeTTS](https://github.com/luoyily/MoeTTS)\n"
            "- [https://github.com/Francis-Komizu/Sovits](https://github.com/Francis-Komizu/Sovits)"
        )
    app.queue(concurrency_count=3).launch(show_api=True, share=True)
    def greet(name):
        return "hahaha " + name + "!"

    demo = gr.Interface(fn=greet, inputs="text", outputs="text")
    demo.launch()