-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
240 lines (215 loc) · 8.96 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
<!doctype html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no">
<title>End-to-end music demixing with pytorch</title>
<link rel="stylesheet" href="css/reveal.css">
<link rel="stylesheet" href="css/theme/lirmm.css">
<link href="fontawesome-free-5.3.1-web/css/all.css" rel="stylesheet">
<!-- <link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.1.0/css/all.css" integrity="sha384-lKuwvrZot6UHsBSfcMvOkWwlCMgc0TaWr+30HWe3a4ltaBwTZhyTEggF5tJv8tbt" crossorigin="anonymous"> -->
<!-- Theme used for syntax highlighting of code -->
<link rel="stylesheet" href="lib/css/github.css">
<!-- Printing and PDF exports -->
<script>
var link = document.createElement( 'link' );
link.rel = 'stylesheet';
link.type = 'text/css';
link.href = window.location.search.match( /print-pdf/gi ) ? 'css/print/pdf.css' : 'css/print/paper.css';
document.getElementsByTagName( 'head' )[0].appendChild( link );
</script>
</head>
<body>
<div class="reveal">
<div class="slides">
<section id="cover" data-background="css/theme/img/background-red.jpg" data-background-repeat="no-repeat" data-background-size="cover" data-background-position="0 0" data-state="no-title-footer no-progressbar has-dark-background">
<h2 id="covertitle">End-to-end music demixing with pytorch</h2>
<p id="coverauthors">
Antoine Liutkus & Fabian-Robert Stöter<br/>
Inria and LIRMM, Montpellier<br/>
<p>
antoine.liutkus@inria.fr<br />
</p>
<p>
fabian-robert.stoter@inria.fr<br />
<i class="fab fa-twitter"></i>faroit
</p>
</p>
<p>
September 16th, 2019
</p>
<p>
<img src="css/theme/img/inria-cover.svg" id="inria" class="logo" alt="">
</section>
<!-- Introduction
In an introductory section, we will motivate the tutorial by explaining how music separation with DNN emerged with
data-driven methods coming from machine-learning or image processing communities. This comes with machine-learning
tricks to make methods work in practice. Meanwhile, many audio processing good practices are often forgotten or not
correctly applied, although they are mandatory for good performance.
• Timeline for DNN music separation
A brief history of the topic in the bigger picture: machine learning on the one hand, source separation on the other
hand.
• Outline
The baseline system: what is the state of the art? Quick literature review
How to implement and make it work? The distinction between training and testing.
An audio demo of the critical impact of engineering for the “same” system.
-->
<!-- INTRODUCTION -->
<section>
<h1>Music Unmixing/Separation</h1>
<img width="45%" style="float:left" src="assets/intro1.png" alt="">
<img class="fragment" width="45%" src="assets/intro2.png" alt="">
</section>
<section>
<h1>Applications</h1>
<img width="60%" style="float:right" src="assets/karaoke.jpg" alt="">
<ul>
<li>Automatic Karaoke</li>
<li>Creative Music Production</li>
<li>Active listening</li>
<li>Upmixing (stereo $\Rightarrow$ 5.1)</li>
<li>Music Education</li>
<li>Pre-processing for MIR</li>
</ul>
</section>
<section>
<h1>Time frequency representations</h1>
<img width="100%" style="margin-top:5%; float:left" src="assets/STFT.svg" alt="">
</section>
<section data-background-transition="none" data-state="no-title-footer" data-background-image="assets/demotrack/mix.jpg">
<!-- <h1><button data-audio="assets/demotrack/mixture.m4a">▶</button></h1> -->
<h1 style="margin-top:5%; color:white">Mixture spectrogram</h1>
</section>
<section data-background-transition="none">
<section data-background-transition="none" data-state="no-title-footer" data-background-image="assets/demotrack/vocals.jpg">
<!-- <h1><button data-audio="assets/demotrack/vocals.m4a">▶</button></h1> -->
<h1 style="margin-top:5%; color:white">Vocals spectrogram</h1>
</section>
<section data-background-transition="none" data-state="no-title-footer" data-background-image="assets/demotrack/drums.jpg">
<!-- <h1><button data-audio="assets/demotrack/drums.m4a">▶</button></h1> -->
<h1 style="margin-top:5%; color:white">Drums spectrogram</h1>
</section>
<section data-background-transition="none" data-state="no-title-footer" data-background-image="assets/demotrack/bass.jpg">
<!-- <h1><button data-audio="assets/demotrack/bass.m4a">▶</button></h1> -->
<h1 style="margin-top:5%; color:white">Bass spectrogram</h1>
</section>
</section>
<section>
<h1>The pytorch <code>Open-unmix</code> (UMX) model</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/UMX1.svg" alt="">
</section>
<section>
<h1>The pytorch <code>Open-unmix</code> (UMX) model</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/UMX2.svg" alt="">
</section>
<section>
<h1>The pytorch <code>Open-unmix</code> (UMX) model</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/UMX.svg" alt="">
</section>
<section>
<h1>The pytorch <code>Open-unmix</code> (UMX) model</h1>
<img style="margin-top:2%;" width="100%" src="assets/UMX1.svg" alt="">
<h2>A state of the art system</h2>
<ul>
<li>MIT-licensed Pytorch implementation</li>
<li><a href=https://github.com/sigsep/open-unmix-pytorch><i class="fab fa-github"></i>/sigsep/open-unmix-pytorch</a> 6.3 dB vocals SDR!</li>
</ul>
<h6 style="margin-top:4%;color:gray">
F. Stöter et al, "Open-Unmix - A reference implementation for audio source separation", JOSS 2019.
</h6>
</section>
<section>
<h1>What <code>open-unmix</code> can achieve</h1>
<iframe src='http://umx-sisec18.s3-website.eu-west-3.amazonaws.com/' style='height:500px;width:100%'></iframe>
</section>
<section>
<h1>Open-unmix: <code>master</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-master_branch1.svg" alt="">
</section>
<section>
<h1>Open-unmix: <code>master</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-master_branch2.svg" alt="">
</section>
<section>
<h1>Open-unmix: <code>master</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-master_branch3.svg" alt="">
</section>
<section>
<h1>Open-unmix: <code>master</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-master_branch4.svg" alt="">
</section>
<section>
<h1>Open-unmix: <code>master</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-master_branch5.svg" alt="">
</section>
<section>
<h1>Our submission to the hackaton!</h1>
<h2>Open-unmix: <code>torchfilters</code></h2>
</section>
<section>
<h1>Open-unmix: <code>master</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-master_branch4.svg" alt="">
</section>
<section>
<h1>Open-unmix: <code>torchfilters</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-torchfilters_branch1.svg" alt="">
</section>
<section>
<h1>Open-unmix: <code>torchfilters</code> branch</h1>
<img style="margin-top:2%; float:left" width="160%" src="assets/open-unmix-torchfilters_branch2.svg" alt="">
</section>
<section>
<h1>Our submission to the hackaton!</h1>
<h2>Open-unmix: <code>torchfilters</code></h2>
<ul>
<li>Stereo Wiener filtering in <code>torch</code></li>
<li class='fragment'>inverse ISTFT with <code>torchaudio</code></li>
<li class='fragment'>End-to-end <code>Separator</code> torch.Module
<img class='fragment' style="margin-top:2%;" width="100%" src="assets/torchhub.png" alt="">
</li>
<li class='fragment'>Computing time for 1mn audio
<ul class='fragment'>Before:
<li>cpu: 31.3s</li>
<li>cuda: 22.5s</li>
</ul>
<ul class='fragment'> After
<li>cpu: 23.4s <font color=red>(25% speedup)</font></li>
<li>cuda: 6.8s <font color=red>(70% speedup)</font></li>
</ul>
</li>
</ul>
</section>
</div>
<div class='footer'>
<img src="css/theme/img/inria-bottom.svg" alt="Logo" />
<div id="middlebox">End-to-end music demixing with pytorch</div>
</div>
</div>
<script src="lib/js/head.min.js"></script>
<script src="js/reveal.js"></script>
<script>
// More info about config & dependencies:
// - https://github.com/hakimel/reveal.js#configuration
// - https://github.com/hakimel/reveal.js#dependencies
Reveal.initialize({
controls: true,
progress: false,
history: true,
center: false,
slideNumber: true,
minScale: 1,
maxScale: 5,
viewDistance: 30,
transition: 'none', //
dependencies: [
{ src: 'plugin/markdown/marked.js' },
{ src: 'plugin/markdown/markdown.js' },
{ src: 'plugin/notes/notes.js', async: true },
{ src: 'node_modules/reveald3/reveald3.js' },
{ src: 'plugin/math-katex/math-katex.js', async: true },
{ src: 'plugin/highlight/highlight.js', async: true, callback: function() { hljs.initHighlightingOnLoad(); } }
]
});
</script>
</body>
</html>