-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathindex.html
305 lines (212 loc) · 9.97 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>
– Pre-trained Text Embeddings for Enhanced Text-to-Speech Synthesis
</title>
<meta name="description" property="og:description" content="Abstract We propose an end-to-end text-to-speech (TTS) synthesis model that explicitly uses information from pre-trained embeddings of the text. Recent work in natural language processing has developed self-supervised representations of text that have proven very effective as pre-training for language understanding tasks. We propose using one such pre-trained representation (BERT) to encode input phrases, as an additional input to a Tacotron2-based sequence-to-sequence TTS model. We hypothesize that the text embeddings contain information about the semantics of the phrase and the importance of each word, which should help TTS systems produce more natural prosody and pronunciation.|">
<meta name="apple-mobile-web-app-title" content="Pre-trained Text Embeddings for Enhanced Text-to-Speech Synthesis">
<link rel="stylesheet" href="/Taco2withBERT/assets/syntax.css">
<link rel="stylesheet" href="/Taco2withBERT/assets/primer-build.css">
<link rel="stylesheet" href="/Taco2withBERT/assets/style.css">
</head>
<body class="bg-gray">
<div id="holy" class="container-lg bg-white h-100">
<div id="header" class="px-1 bg-white">
<nav class="UnderlineNav UnderlineNav--right px-2">
<a class="UnderlineNav-actions muted-link h2" href="https://kan-bayashi.github.io/Taco2withBERT/">
Pre-trained Text Embeddings for Enhanced Text-to-Speech Synthesis
</a>
</nav>
</div>
<div role="main" id="main" class="holy-main markdown-body px-4 bg-white">
<div class="Subhead">
<div class="Subhead-heading">
<div class="h1 mt-3 mb-1"></div>
</div>
<div class="Subhead-description">
<div class="float-md-right">
<span title="Lastmod: 2019-07-01. Published at: 2019-07-01.">
Published: 2019-07-01
</span>
</div>
</div>
</div>
<article>
<section class="pb-6 mb-3 border-bottom">
<h2 id="abstract">Abstract</h2>
<p>We propose an end-to-end text-to-speech (TTS) synthesis model that explicitly uses information from pre-trained
embeddings of the text. Recent work in natural language processing has developed self-supervised representations
of text that have proven very effective as pre-training for language understanding tasks. We propose using one
such pre-trained representation (BERT) to encode input phrases, as an additional input to a Tacotron2-based
sequence-to-sequence TTS model. We hypothesize that the text embeddings contain information about the semantics
of the phrase and the importance of each word, which should help TTS systems produce more natural prosody and
pronunciation. We conduct subjective listening tests of our proposed models using the 24-hour LJSpeech corpus,
finding that they improve mean opinion scores modestly but significantly over a baseline TTS model without
pre-trained text embedding input.</p>
<h2 id="generated-examples">Generated examples</h2>
<h3 id="baseline">Baseline</h3>
<p><audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0075.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0090.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0094.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0098.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<h3 id="phrase-level-model">Phrase-level model</h3>
<p><audio controls="controls" >
<source src="/Taco2withBERT/wav/sentence-based-bert/LJ050-0075.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/sentence-based-bert/LJ050-0090.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/sentence-based-bert/LJ050-0094.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/sentence-based-bert/LJ050-0098.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<h3 id="subword-level-model">Subword-level model</h3>
<p><audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0075.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0090.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0094.wav" autoplay/>
Your browser does not support the audio element.
</audio>
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0098.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<h2 id="examples-from-a-b-forced-choice-with-high-agreement">Examples from A/B forced choice with high agreement</h2>
<h3 id="subword-example-preferred">Subword example preferred</h3>
<p><code>result in some degree of interference with the personal liberty of those involved.</code></p>
<p><strong>Subword</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0073.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><strong>Baseline</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0073.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><code>In June 1964, the Secret Service sent to a number of Federal law enforcement and intelligence agencies</code></p>
<p><strong>Subword</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0078.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><strong>Baseline</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0078.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><code>determination to use a means, other than legal or peaceful, to satisfy his grievance, end quote, within the meaning of the new criteria.</code></p>
<p><strong>Subword</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0098.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><strong>Baseline</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0098.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<h3 id="baseline-example-preferred">Baseline example preferred</h3>
<p><code>it has obtained the services of outside consultants, such as the Rand Corporation,</code></p>
<p><strong>Subword</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0046.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><strong>Baseline</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0046.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><code>and from a specialist in psychiatric prognostication at Walter Reed Hospital.</code></p>
<p><strong>Subword</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/subword-based-bert/LJ050-0049.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<p><strong>Baseline</strong> :
<audio controls="controls" >
<source src="/Taco2withBERT/wav/baseline/LJ050-0049.wav" autoplay/>
Your browser does not support the audio element.
</audio></p>
<h2 id="citation">Citation</h2>
<pre><code>@inproceedings{hayashi2019pretrained,
title={Pre-trained Text Embeddings for Enhanced Text-to-Speech Synthesis},
author={Hayashi, Tomoki and Watanabe, Shinji and Toda, Tomoki and Takeda, Kazuya and Toshniwal, Shubham and Livescu, Karen},
booktitle={Interspeech 2019 (Accepted)},
year={2019}
}
</code></pre>
</section>
<section>
</section>
</article>
</div>
<div id="side" class="pr-1 bg-white">
<aside class="pr-3">
<div id="toc" class="Box Box--blue mb-3">
<b></b>
<nav id="TableOfContents">
<ul>
<li>
<ul>
<li><a href="#abstract">Abstract</a></li>
<li><a href="#generated-examples">Generated examples</a>
<ul>
<li><a href="#baseline">Baseline</a></li>
<li><a href="#phrase-level-model">Phrase-level model</a></li>
<li><a href="#subword-level-model">Subword-level model</a></li>
</ul></li>
<li><a href="#examples-from-a-b-forced-choice-with-high-agreement">Examples from A/B forced choice with high agreement</a>
<ul>
<li><a href="#subword-example-preferred">Subword example preferred</a></li>
<li><a href="#baseline-example-preferred">Baseline example preferred</a></li>
</ul></li>
<li><a href="#citation">Citation</a></li>
</ul></li>
</ul>
</nav>
</div>
<div>
</div>
</aside>
</div>
<div id="footer" class="pt-2 pb-3 bg-white text-center">
<span class="text-small text-gray">
Powered by the
<a href="https://github.com/qqhann/hugo-primer" class="link-gray-dark">Hugo-Primer</a> theme for
<a href="https://gohugo.io" class="link-gray-dark">Hugo</a>.
</span>
</div>
</div>
<script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
<script type="text/x-mathjax-config">MathJax.Hub.Config({ tex2jax: { inlineMath: [['$','$'], ['\\(','\\)']] } });</script>
</body>
</html>