-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
328 lines (267 loc) · 12.6 KB
/
index.html
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
328
<!doctype html>
<html>
<head>
<!-- BOOTSTRAP CORE STYLE -->
<link rel="stylesheet" href="https://stackpath.bootstrapcdn.com/bootstrap/4.5.0/css/bootstrap.min.css" integrity="sha384-9aIt2nRpC12Uk9gS9baDl411NQApFmC26EwAOH8WgZl5MYYxFfc+NcPb1dKGj7Sk" crossorigin="anonymous">
<!-- GOOGLE FONT -->
<link href='https://fonts.googleapis.com/css?family=Open+Sans' rel='stylesheet' type='text/css' crossorigin="anonymous" />
<!-- BOOTSTRAP JS -->
<script src="https://code.jquery.com/jquery-3.5.1.slim.min.js" integrity="sha384-DfXdz2htPH0lsSSs5nCTpuj/zy4C+OGpamoFVy38MVBnE+IbbVYUew+OrCXaRkfj" crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/popper.js@1.16.0/dist/umd/popper.min.js" integrity="sha384-Q6E9RHvbIyZFJoft+2mJbHaEWldlvI9IOYy5n3zV9zzTtmI3UksdQRVvoxMfooAo" crossorigin="anonymous"></script>
<script src="https://stackpath.bootstrapcdn.com/bootstrap/4.5.0/js/bootstrap.min.js" integrity="sha384-OgVRvuATP1z7JjHLkuOU7Xw704+h835Lr+6QL9UvYjZE3Ipu6Tp75j7Bh/kR0JKI" crossorigin="anonymous"></script>
<!-- CUSTOM STYLE CSS -->
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1" />
<!-- Page description -->
<!-- Page keywords -->
<!-- Page author -->
<meta name="author" content="Masaya Kawamura" />
<!-- Site title -->
<title>Demonstration of Lightweight and High-Fidelity End-to-End Text-to-Speech with Multi-Band Generation and Inverse Short-Time Fourier Transform - Masaya Kawamura</title>
</head>
<body>
<section class="jumbotron text-center">
<div class="container">
<h1 class="jumbotron-heading">Lightweight and High-Fidelity End-to-End Text-to-Speech with Multi-Band Generation and Inverse Short-Time Fourier Transform</h1>
<p class="lead">
Masaya Kawamura<sup>1*</sup>, Yuma Shirahata<sup>2</sup>, Ryuichi Yamamoto<sup>2</sup>, Kentaro Tachibana<sup>2</sup>
</p>
<p class="lead">
<sup>1</sup>The University of Tokyo, Tokyo, Japan<br>
<sup>2</sup>LINE Corp., Japan.<br>
<sup>*</sup>Work performed during an internship at LINE corporation<br><br>
</p>
<p class="lead">
Accepted to ICASSP 2023<br>
[<a href="https://arxiv.org/abs/2210.15975">Paper</a>] [<a href="https://github.com/MasayaKawamura/MB-iSTFT-VITS">Code</a>]
</p>
</div>
</section>
<p><link rel="stylesheet" type="text/css" href="style.css"></p>
<div class="container">
<div class="row"><h2>Abstract</h2></div>
<div class="row">
<p class="lead">
We propose a lightweight end-to-end text-to-speech model using multi-band generation
and inverse short-time Fourier transform. Our model is based on VITS, a high-quality
end-to-end text-to-speech model, but adopts two changes for more efficient inference: 1) the most
computationally expensive component is partially replaced with a simple inverse short-time Fourier
transform, and 2) multi-band generation, with fixed or trainable synthesis filters, is used to generate
waveforms. Unlike conventional lightweight models, which employ optimization or knowledge distillation separately
to train two cascaded components, our method enjoys the full benefits of end-to-end optimization. Experimental results
show that our model synthesized speech as natural as that synthesized by VITS, while achieving a real-time factor of
0.066 on an Intel Core i7 CPU, 4.1 times faster than VITS. Moreover, a smaller version of the model significantly
outperformed a lightweight baseline model with respect to both naturalness and inference speed.
Code and audio samples are available from <a href="https://github.com/MasayaKawamura/MB-iSTFT-VITS">https://github.com/MasayaKawamura/MB-iSTFT-VITS</a>.
</p>
</div>
<div align="center">
<img src="images/proposed_figure.png" width="100%">
<figcaption>Architecture of multi-band iSTFT VITS and multi-stream iSTFT VITS.</figcaption>
</div>
<div class="row"><h2>Demo</h2></div>
<div class="row">
<p class="lead">
This is an accompanying page and includes some examples of the synthesized speech obtained with the proposed and conventional methods.
The groundtruth audio clips are from the LJ speech dataset [1].
These clips are not included in the training data.
</p>
</div>
</div>
<p><br></p>
<table style="table-layout: fixed;">
<thead>
<tr>
<th style="vertical-align:middle" style="text-align: center" class="td_fixed">Model</th>
<th style="text-align: center" >Text: Weedon and Lecasser to twelve <br>and six months respectively in Coldbath Fields.</th>
<th style="text-align: center" >Text: There among the ruins <br>they still live in the same kind of houses,</th>
<th style="text-align: center" >Text: Mrs. De Mohrenschildt thought that Oswald,</th>
<th style="text-align: center" >Text: carbohydrates (starch, cellulose) and fats.</th>
</tr>
</thead>
<tbody>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed">Ground truth</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/GT_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/GT_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/GT_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/GT_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed">VITS [2]</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/VITS_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/VITS_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/VITS_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/VITS_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed">Nix-TTS [3]</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/NixTTS_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/NixTTS_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/NixTTS_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/NixTTS_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed">iSTFT-VITS</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/VITS+iSTFTNet_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/VITS+iSTFTNet_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/VITS+iSTFTNet_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/VITS+iSTFTNet_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed"><u>MB-iSTFT-VITS</u></td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/VITS+Multiband_iSTFTNet_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/VITS+Multiband_iSTFTNet_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/VITS+Multiband_iSTFTNet_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/VITS+Multiband_iSTFTNet_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed"><u>MS-iSTFT-VITS</u></td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/VITS+Multistream_iSTFTNet_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/VITS+Multistream_iSTFTNet_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/VITS+Multistream_iSTFTNet_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/VITS+Multistream_iSTFTNet_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed">Mini-VITS</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/MiniVITS_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/MiniVITS_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/MiniVITS_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/MiniVITS_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed">Mini-iSTFT-VITS</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/MiniVITS+iSTFTNet_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/MiniVITS+iSTFTNet_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/MiniVITS+iSTFTNet_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/MiniVITS+iSTFTNet_LJ026-0054.mp3">
</audio>
</td>
</tr>
<tr>
<td style="vertical-align:middle" style="text-align: center" class="td_fixed"><u>Mini-MB-iSTFT-VITS</u></td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ012-0015/MiniVITS+Multiband_iSTFTNet_LJ012-0015.mp3">
</audio>
</td >
<td >
<audio controls preload="metadata" src="demo/Sample_LJ028-0410/MiniVITS+Multiband_iSTFTNet_LJ028-0410.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ045-0096/MiniVITS+Multiband_iSTFTNet_LJ045-0096.mp3">
</audio>
</td>
<td >
<audio controls preload="metadata" src="demo/Sample_LJ026-0054/MiniVITS+Multiband_iSTFTNet_LJ026-0054.mp3">
</audio>
</td>
</tr>
</tbody>
</table>
<div class="container">
<h1>References</h1>
<div>
[1] K. Ito, "The LJ speech dataset", <a href= "https://keithito.com/LJ-Speech-Dataset/" >https://keithito.com/LJ-Speech-Dataset/</a>, 2017.
<br>
[2] J. Kim, J. Kong, J. Son, "Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech," in <em>Proc. ICML</em>, 2021, pp. 5530-5540.
<br>
[3] R. Chevi, R. E. Prasojo, A. F. Aji, A. Tjandra, S. Sakti, "Nix-TTS: Lightweight and end-to-end text-to-speech via module-wise distillation", in <em>Proc. SLT</em>, 2023, pp. 970–976.
</div>
</div>
</body>
</html>