-
Notifications
You must be signed in to change notification settings - Fork 6
/
sltu.html
102 lines (86 loc) · 4.69 KB
/
sltu.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
<!DOCTYPE html>
<html>
<head>
<meta name="description" content="Kaldi ASR Tutorial for SLTU'16 Participants"/>
<meta charset="UTF-8">
<link rel="icon" type="image/png" href="/kaldi_ico.png"/>
<link rel="stylesheet" type="text/css" href="/style.css"/>
<title>Kaldi ASR Tutorial for SLTU'16 Participants</title>
</head>
<body>
<div class="container">
<div id="centeredContainer">
<div id="headerBar">
<div id="headerLeft"> <a href="http://kaldi-asr.org"><image id="logoImage" src="/kaldi_text_and_logo.png"></a> </div>
<div id="headerRight"> <image id="logoImage" src="/kaldi_logo.png"> </div>
<!-- <h2 class="kaldiStyle"> Kaldi </h2> -->
</div>
<hr>
<div id="topBar">
<a class="topButtons" href="/index.html">Home</a>
<a class="topButtons" href="/doc/">Documentation</a>
<a class="myTopButton" href="/forums.html">Help!</a>
<a class="topButtons" href="/models.html">Models</a>
</div>
<hr>
<div id="rightCol">
<div class = "contact_info">
<div class="contactTitle">Team</div>
Sanjeev Khudanpur<br/>
Jan "Yenda" Trmal<br/>
Vijayaditya Peddinti<br/>
Sarah Samson Juan<br/>
Dessi Puji Lestari<br/>
</div>
</div>
<div id="mainContent">
<div class= "container" >
<h3 class="kaldiStyle">BUILDING SPEECH RECOGNITION SYSTEMS WITH THE KALDI TOOLKIT</h3>
<h4 class="kaldiStyle">Information for participants</h4>
The current version of the <a href="downloads/tutorial-materials/sltu-data/2016-05-SLTU-Workshop-2.pdf">tutorial slides</a>. <b>Updated on 2016-05-10</b>.
<h4 class="kaldiStyle">Logging onto a machine</h4>
Using the provided SSH <a href="downloads/tutorial-materials/sltu-data/sltu-public.pem.txt">private key</a>, login on a assigned machine.
Using the command line SSH (linux/macosX), this can be done using the following command:
<pre>
ssh -i sltu-public.pem.txt ubuntu@<machine-name>
</pre>
where <tt><machine-name></tt> is the address of the machine with the same number you've been assigned.
Please note that after downloading, you might need to change the access rights to the file <tt>sltu-public.pem.txt</tt>
to allow only the current user to read the file (it's a security precaution built-in into OpenSSH clients).
This can be done using the following command
<pre>
chmod 600 sltu-public.pem.txt
</pre>
<h4 class="kaldiStyle">Running the recipe</h4>
Once you've log in, run the command <tt>screen</tt>. That command ensures that even if the connection to the machine is dropped, the scripts will still keep running, so you won't have to run everything from scratch.
<p/> The recipe is in <tt>~/kaldi/egs/iban/s5</tt>. Everything should be set up correctly for you, so that you can go ahead and run the script <tt>run.sh</tt>. We will use <tt>tee</tt> to keep the console output in the log file for a future reference.
<pre>
~/kaldi/egs/iban/s5 $ ./run.sh 2>&1 | tee run.log
</pre>
It will take a two or three hours to finish.
After the script finishes, you can run another script, providing training procedure for TDNN acoustic models. That script is <tt>./local/nnet3/run_tdnn.sh</tt>.
It is inteded to be run from the same directory in which the script <tt>run.sh</tt> lies, i.e. should be executed as
<pre>
~/kaldi/egs/iban/s5 $ ./local/nnet3/run_tdnn.sh 2>&1 | tee run-tdnn.log
</pre>
<h4 class="kaldiStyle">List of machines</h4>
<b>The machines are now (as of 2016-05-10) offline.</b> The Iban recipe is include into the Kaldi standard egs. The data are publicly available (see <a href="http://www.openslr.org/24/">http://www.openslr.org/24/</a>). You can train the system using your own local machine/cluster.
<p/>
We have published two AWS AMI images "v4 IBAN recipe/Ubuntu 14.04LTS" (<tt>AMI-ID ami-7b688916</tt>) and "v5 IBAN recipe/Ubuntu 14.04LTS" (<tt>AMI-ID ami-6945a804</tt>). The difference between the two is tht "v4" has a complete setup, but does not contain the files created during training. The "v5" contains fully trained and decoded system.
</div>
</div>
</div>
<script type="text/javascript" src="http://ajax.googleapis.com/ajax/libs/jquery/1.4.2/jquery.min.js"></script>
<div style="clear: both"></div>
<div id="footer">
<p>
<a href="http://jigsaw.w3.org/css-validator/check/referer">
<img style="border:0;width:88px;height:31px"
src="http://jigsaw.w3.org/css-validator/images/vcss-blue"
alt="Valid CSS!" />
</a>
</p>
</div>
</div>
</body>
</html>