-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
327 lines (261 loc) · 11.8 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
<!DOCTYPE html>
<!--[if lt IE 7]> <html class="no-js ie6" lang="en"> <![endif]-->
<!--[if IE 7]> <html class="no-js ie7" lang="en"> <![endif]-->
<!--[if IE 8]> <html class="no-js ie8" lang="en"> <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en"> <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1">
<title>Automatic Emotional Speech Recognition</title>
<meta name="description" content="A jQuery library for modern HTML presentations">
<meta name="author" content="Caleb Troughton">
<meta name="viewport" content="width=1024, user-scalable=no">
<!-- Core and extension CSS files -->
<link rel="stylesheet" href="core/deck.core.css">
<link rel="stylesheet" href="extensions/goto/deck.goto.css">
<link rel="stylesheet" href="extensions/menu/deck.menu.css">
<link rel="stylesheet" href="extensions/navigation/deck.navigation.css">
<link rel="stylesheet" href="extensions/status/deck.status.css">
<link rel="stylesheet" href="extensions/hash/deck.hash.css">
<link rel="stylesheet" href="extensions/scale/deck.scale.css">
<!-- Style theme. More available in /themes/style/ or create your own. -->
<link rel="stylesheet" href="themes/style/swiss.css">
<!-- Transition theme. More available in /themes/transition/ or create your own. -->
<link rel="stylesheet" href="themes/transition/horizontal-slide.css">
<!-- Syntax highligher -->
<link rel="stylesheet" href="http://yandex.st/highlightjs/7.3/styles/default.min.css">
<script src="http://yandex.st/highlightjs/7.3/highlight.min.js"></script>
<script>hljs.initHighlightingOnLoad();</script>
<script src="modernizr.custom.js"></script>
</head>
<body class="deck-container">
<!-- Begin slides -->
<section class="slide" id="title-slide">
<h1>Automatic Emotional Speech Recognition</h1>
</section>
<section class="slide" id="outline">
<h2>Outline</h2>
<ul>
<li>Context</li>
<li>What it is and what it's not</li>
<li>Theories</li>
<li>How to build a recognition system</li>
<li>Some references</li>
</ul>
</section>
<section class="slide" id="">
<h1>Context</h1>
</section>
<section class="slide" id="context">
<h2>Affective Computing</h2>
<p>New research field launched by Pr. Rosalind Picard (MIT Media Lab) in 1995</p>
<p>In short, machines that can recognize, interpret and simulate human emotions.</p>
<p>Why? → non-verbal cues are essential to human communication.</p>
<div class="slide">
<h2>Applications</h2>
<p>Everywhere when there's a need for HMI.</p>
<p>Some specific applications:</p>
<ul>
<li>Stress detection for airline pilots</li>
<li>Social and empathic training for autistic persons</li>
<li>Feedback loop in therapy tools</li>
<li>Anger detection for call centers</li>
<li>Videogames (automatic difficulty adaptation...)</li>
</ul>
</div>
</section>
<section class="slide">
<h2>Emotional Speech is just a tiny fraction of this</h2>
<ul>
<li>Facial expressions</li>
<li>Gestures</li>
<li>Emotional Speech Synthesis</li>
<li>Convincing virtual emotional characters / robots</li>
<li>...</li>
</ul>
<p>NB: we restrict ourselves to the detection of emotion from voice without any understanding of linguistics (only prosody)</p>
</section>
<section class="slide" id="what-it-is">
<h2>What it is</h2>
<ul>
<li>a booming research field: teams in Europe, USA + Canada, Asia...</li>
<li>a work in progress: current systems are far from perfect</li>
<li>an attempt to translate into technology years of previous research in human sciences</li>
<li>something you can tinker with using available tools and databases</li>
</ul>
<section class="slide" id="">
<h2>What it's not</h2>
<ul>
<li>an off-the-self technology that you can easily integrate in a commercial product</li>
<li>lie detection technology (already exists...)</li>
<li>terrorists detection technology (although <a href="http://spectrum.ieee.org/computing/embedded-systems/loser-bad-vibes/0">some would like it to be...</a>)</li>
</ul>
<p>It raises a number of questions for sure (privacy, misuse...). Researchers are <a href="http://www.cnrs.fr/comets/IMG/pdf/02-comstic.pdf">actually thinking</a> about this (CNRS ethics comitee report, 2009, pp 20, 70)</p>
</section>
</section>
<section class="slide" id="">
<h1>Theories</h1>
</section>
<section class="slide" id="theory">
<h2>Long story</h2>
<p>Emotions have been studied for a long time in human sciences, psychology and medicine</p>
<p>Nobody agrees on emotion theory, different perspective exist:</p>
<ul>
<li>Darwin (1872): emotions are an evolutionary construction with important survival benefits</li>
<li>Jameson (end 19th century): strong link between bodily functions and reactions and emotions</li>
<li>End of 20th century: social constructivism</li>
<li>Ekman (1984): work on universal emotions</li>
</ul>
<section class="slide">
<p>Darwin's book "The Expression of the Emotions in Man and Animals" is probably the most influencial book</p>
<p>Look at it! It's <a href="http://www.gutenberg.org/ebooks/1227">freely downloadable</a> and has weird pictures of people being electrocuted (Duchenne's work)</p>
<img src="images/duchenne.jpg"/>
</section>
</section>
<section class="slide" id="">
<h1>How-to</h1>
</section>
<section class="slide" id="how-to-data">
<h2>Data</h2>
<section class="slide">
<p>Systems try to imitate human perception</p>
<p>→ You need to know what it looks like</p>
<p>→ In practice, you use labelled data.</p>
</section>
<section class="slide">
<h3>You need to:</h3>
<ul>
<li>find data (usually HARD)</li>
<li>make it tidy (audio segmentation, BORING)</li>
<li>label it (always LONG)</li>
</ul>
<p>Problem: this yields very little results and is very expensive</p>
</section>
<section class="slide">
<h3>What people do:</h3>
<ul>
<li>use very few data (BAD for learning and system performance)</li>
<li>use actors in lab settings (BAD because not representative)</li>
<li>use Amazon's Mechanical Turk to annotate (seems quite OK, very cost-effective)</li>
</ul>
<p>There are some available databases, usually for research purposes (<a href="http://emotion-research.net/wiki/Databases">link</a>)</p>
</section>
</section>
<section class="slide" id="how-to-features">
<h2>Features extraction</h2>
<section class="slide">
<h3>Basics features:</h3>
<ul>
<li>frequency (pitch, MFCC...)</li>
<li>energy</li>
<li>voice quality (jitter, shimmer...)</li>
<li>sometimes rhythm features</li>
</ul>
<img src='images\praat.jpg'/>
</section>
<section class="slide">
<h3>Statistics/functionals:</h3>
<ul>
<li>mean, std, kurtosis, skewness...</li>
<li>median, quartiles</li>
<li>first and second order temporal derivatives</li>
<li>linear prediction coefficients</li>
<li>frequential analysis</li>
<li>mad combinations of the aboves...</li>
</ul>
<p>A good library for audio features extraction: <a href="http://opensmile.sourceforge.net/">openSMILE</a></p>
</section>
</section>
<section class="slide" id="how-to-machine-learning">
<h2>Automatic learning</h2>
<section class="slide">
<p>We use algorithms that can handle having so much dimensions in the data</p>
<ul>
<li>SVMs are popular</li>
<li>Neural Networks are making a come-back with Deep-Belief Networks</li>
<li>a little bit of decision trees here and there (Random Forests, C4.5...), GMMs...</li>
<li>interestingly, not a lot of temporal modelling (HMMs)</li>
</ul>
</section>
<section class="slide">
<p>Increasingly some forms of Feature Selection are used:</p>
<ul>
<li>most of the time simple filter approaches</li>
<li>sometimes through regularization, inside the training process</li>
<li>sometimes with more sophisticated methods (SFFS)</li>
</ul>
</section>
</section>
<section class="slide" id="how-to-machine-learning-advice">
<h2>Advice</h2>
<ul>
<li>Make yourself a small database that suits your application or try experimenting with a very simple available one (<a href="http://www.expressive-speech.net/">Berlin</a> for instance)</li>
<li>Use simple models first (linear SVMs for instance)</li>
<li>Normalize your data!</li>
<li>Not too many features if you don't have a cluster available... a few dozen is good</li>
</ul>
<p>A widely-used SVM library: <a href="http://www.csie.ntu.edu.tw/~cjlin/libsvm/">libSVM</a></p>
</section>
<section class="slide" id="how-to-real-system">
<h2>Building a prototype</h2>
<p>Not too complicated when you use available tools and databases</p>
<p>Example: PartyMixer, a project from <a href="http://www.musichackparis.org/hacks">MusicHackParis</a>, a few hours</p>
</section>
<section class="slide" id="">
<h1>References</h1>
</section>
<section class="slide" id="some-references">
<h2>Scientific papers / courses / books</h2>
<ul>
<li><a href="http://cvrr.ucsd.edu/ece285/papers/Zeng_SurveyAffectRecognition.pdf">A good review on Affect Recognition</a></li>
<li>A very good <a href="https://www.youtube.com/playlist?list=PLD63A284B7615313A">Caltech course</a> on Machine Learning (video, quite hard)</li>
<li><a href="http://cs.gmu.edu/~sean/book/metaheuristics/">Book on metaheuristics and optimization</a></li>
</ul>
<h2>Tools</h2>
<ul>
<li><a href="http://trans.sourceforge.net/en/presentation.php">Transcriber</a>, for audio labelling</li>
<li><a href="http://www.cs.waikato.ac.nz/ml/weka/">Weka</a>, a complete machine learning toolkitlibSVM</li>
</section>
<section class="slide" id="questions">
<h2>Questions?</h2>
<section class="slide">
<h3>Drop me a line: <a href="mailto:[email protected]">[email protected]</a></h3>
</section>
</section>
<!-- deck.navigation snippet -->
<a href="#" class="deck-prev-link" title="Previous">←</a>
<a href="#" class="deck-next-link" title="Next">→</a>
<!-- deck.status snippet -->
<p class="deck-status">
<span class="deck-status-current"></span>
/
<span class="deck-status-total"></span>
</p>
<!-- deck.goto snippet -->
<form action="." method="get" class="goto-form">
<label for="goto-slide">Go to slide:</label>
<input type="text" name="slidenum" id="goto-slide" list="goto-datalist">
<datalist id="goto-datalist"></datalist>
<input type="submit" value="Go">
</form>
<!-- deck.hash snippet -->
<a href="." title="Permalink to this slide" class="deck-permalink">#</a>
<!-- Grab CDN jQuery, with a protocol relative URL; fall back to local if offline -->
<script src="//ajax.aspnetcdn.com/ajax/jQuery/jquery-1.7.min.js"></script>
<script>window.jQuery || document.write('<script src="jquery-1.7.min.js"><\/script>')</script>
<!-- Deck Core and extensions -->
<script src="core/deck.core.js"></script>
<script src="extensions/hash/deck.hash.js"></script>
<script src="extensions/menu/deck.menu.js"></script>
<script src="extensions/goto/deck.goto.js"></script>
<script src="extensions/status/deck.status.js"></script>
<script src="extensions/navigation/deck.navigation.js"></script>
<script src="extensions/scale/deck.scale.js"></script>
<!-- Initialize the deck -->
<script>
$(function() {
$.deck('.slide');
});
</script>
</body>
</html>