forked from sooftware/kospeech
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Speech Transformer.html
573 lines (441 loc) · 54.1 KB
/
Speech Transformer.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
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
<!DOCTYPE html>
<html class="writer-html5" lang="en" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Speech Transformer — KoSpeech latest documentation</title>
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<!--[if lt IE 9]>
<script src="_static/js/html5shiv.min.js"></script>
<![endif]-->
<script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<script src="_static/language_data.js"></script>
<script type="text/javascript" src="_static/js/theme.js"></script>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Jasper" href="Jasper.html" />
<link rel="prev" title="RNN Transducer" href="RNN%20Transducer.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="index.html" class="icon icon-home"> KoSpeech
</a>
<div class="version">
latest
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<p class="caption"><span class="caption-text">GETTING STARTED</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html">What’s New</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#note">Note</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#kospeech-open-source-toolkit-for-end-to-end-korean-speech-recognition-paper"><em><strong>KoSpeech: Open-Source Toolkit for End-to-End Korean Speech Recognition [Paper]</strong></em></a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#supported-models">Supported Models</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#pre-processed-transcripts">Pre-processed Transcripts</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#introduction">Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#roadmap">Roadmap</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#installation">Installation</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#get-started">Get Started</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#troubleshoots-and-contributing">Troubleshoots and Contributing</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/intro.html#citation">Citation</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/Preparation.html">KsponSpeech-preprocess</a></li>
</ul>
<p class="caption"><span class="caption-text">ARCHITECTURE</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="Interface.html">Interface</a></li>
<li class="toctree-l1"><a class="reference internal" href="Deep%20Speech%202.html">Deep Speech 2</a></li>
<li class="toctree-l1"><a class="reference internal" href="Listen%20Attend%20Spell.html">Listen, Attend and Spell</a></li>
<li class="toctree-l1"><a class="reference internal" href="RNN%20Transducer.html">RNN Transducer</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Speech Transformer</a><ul>
<li class="toctree-l2"><a class="reference internal" href="#module-kospeech.models.transformer.model">Transformer</a></li>
<li class="toctree-l2"><a class="reference internal" href="#module-kospeech.models.transformer.encoder">Encoder</a></li>
<li class="toctree-l2"><a class="reference internal" href="#module-kospeech.models.transformer.decoder">Decoder</a></li>
<li class="toctree-l2"><a class="reference internal" href="#module-kospeech.models.transformer.sublayers">Sublayers</a></li>
<li class="toctree-l2"><a class="reference internal" href="#module-kospeech.models.transformer.embeddings">Embeddings</a></li>
<li class="toctree-l2"><a class="reference internal" href="#module-kospeech.models.transformer.mask">Mask</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="Jasper.html">Jasper</a></li>
<li class="toctree-l1"><a class="reference internal" href="Conformer.html">Conformer</a></li>
<li class="toctree-l1"><a class="reference internal" href="Modules.html">Modules</a></li>
</ul>
<p class="caption"><span class="caption-text">LIBRARY REFERENCE</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="Checkpoint.html">Checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="Criterion.html">Criterion</a></li>
<li class="toctree-l1"><a class="reference internal" href="Data.html">Data</a></li>
<li class="toctree-l1"><a class="reference internal" href="Decode.html">Decode</a></li>
<li class="toctree-l1"><a class="reference internal" href="Evaluator.html">Evaluator</a></li>
<li class="toctree-l1"><a class="reference internal" href="Optim.html">Optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="Learning%20Rate%20Schedulers.html">Learning Rate Scheduler</a></li>
<li class="toctree-l1"><a class="reference internal" href="Trainer.html">Trainer</a></li>
<li class="toctree-l1"><a class="reference internal" href="Vocabs.html">Vocabs</a></li>
<li class="toctree-l1"><a class="reference internal" href="Etc.html">Etc</a></li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
<nav class="wy-nav-top" aria-label="top navigation">
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="index.html">KoSpeech</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="wy-breadcrumbs">
<li><a href="index.html" class="icon icon-home"></a> »</li>
<li>Speech Transformer</li>
<li class="wy-breadcrumbs-aside">
<a href="_sources/Speech Transformer.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<div class="section" id="speech-transformer">
<h1>Speech Transformer<a class="headerlink" href="#speech-transformer" title="Permalink to this headline">¶</a></h1>
<div class="section" id="module-kospeech.models.transformer.model">
<span id="transformer"></span><h2>Transformer<a class="headerlink" href="#module-kospeech.models.transformer.model" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="kospeech.models.transformer.model.SpeechTransformer">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.model.</code><code class="sig-name descname">SpeechTransformer</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">input_dim</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span></em>, <em class="sig-param"><span class="n">num_classes</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span></em>, <em class="sig-param"><span class="n">extractor</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.9)">str</a></span></em>, <em class="sig-param"><span class="n">num_encoder_layers</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">12</span></em>, <em class="sig-param"><span class="n">num_decoder_layers</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">6</span></em>, <em class="sig-param"><span class="n">encoder_dropout_p</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)">float</a></span> <span class="o">=</span> <span class="default_value">0.2</span></em>, <em class="sig-param"><span class="n">decoder_dropout_p</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)">float</a></span> <span class="o">=</span> <span class="default_value">0.2</span></em>, <em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">d_ff</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">2048</span></em>, <em class="sig-param"><span class="n">pad_id</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">0</span></em>, <em class="sig-param"><span class="n">sos_id</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">1</span></em>, <em class="sig-param"><span class="n">eos_id</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">2</span></em>, <em class="sig-param"><span class="n">num_heads</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">8</span></em>, <em class="sig-param"><span class="n">joint_ctc_attention</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)">bool</a></span> <span class="o">=</span> <span class="default_value">False</span></em>, <em class="sig-param"><span class="n">max_length</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">400</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/model.html#SpeechTransformer"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.model.SpeechTransformer" title="Permalink to this definition">¶</a></dt>
<dd><p>A Speech Transformer model. User is able to modify the attributes as needed.
The model is based on the paper “Attention Is All You Need”.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>input_dim</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – dimension of input vector</p></li>
<li><p><strong>num_classes</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – number of classification</p></li>
<li><p><strong>extractor</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.9)"><em>str</em></a>) – type of CNN extractor (default: vgg)</p></li>
<li><p><strong>num_encoder_layers</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a><em>, </em><em>optional</em>) – number of recurrent layers (default: 12)</p></li>
<li><p><strong>num_decoder_layers</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a><em>, </em><em>optional</em>) – number of recurrent layers (default: 6)</p></li>
<li><p><strong>encoder_dropout_p</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)"><em>float</em></a><em>, </em><em>optional</em>) – dropout probability of encoder (default: 0.2)</p></li>
<li><p><strong>decoder_dropout_p</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)"><em>float</em></a><em>, </em><em>optional</em>) – dropout probability of decoder (default: 0.2)</p></li>
<li><p><strong>d_model</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – dimension of model (default: 512)</p></li>
<li><p><strong>d_ff</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – dimension of feed forward net (default: 2048)</p></li>
<li><p><strong>pad_id</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – identification of <PAD_token> (default: 0)</p></li>
<li><p><strong>sos_id</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – identification of <SOS_token> (default: 1)</p></li>
<li><p><strong>eos_id</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – identification of <EOS_token> (default: 2)</p></li>
<li><p><strong>num_heads</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a>) – number of attention heads (default: 8)</p></li>
<li><p><strong>max_length</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)"><em>int</em></a><em>, </em><em>optional</em>) – max decoding step (default: 400)</p></li>
<li><p><strong>joint_ctc_attention</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)"><em>bool</em></a><em>, </em><em>optional</em>) – flag indication joint ctc attention or not (default: False)</p></li>
</ul>
</dd>
</dl>
<dl class="simple">
<dt>Inputs: inputs, input_lengths, targets, teacher_forcing_ratio</dt><dd><ul class="simple">
<li><p><strong>inputs</strong> (torch.Tensor): tensor of sequences, whose length is the batch size and within which
each sequence is a list of token IDs. This information is forwarded to the encoder.</p></li>
<li><p><strong>input_lengths</strong> (torch.Tensor): tensor of sequences, whose contains length of inputs.</p></li>
<li><p><strong>targets</strong> (torch.Tensor): tensor of sequences, whose length is the batch size and within which
each sequence is a list of token IDs. This information is forwarded to the decoder.</p></li>
</ul>
</dd>
</dl>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p><p>(Tensor, Tensor, Tensor)</p>
<ul class="simple">
<li><p>predicted_log_probs (torch.FloatTensor): Log probability of model predictions.</p></li>
<li><p>encoder_output_lengths: The length of encoder outputs. <code class="docutils literal notranslate"><span class="pre">(batch)</span></code></p></li>
<li><dl class="simple">
<dt>encoder_log_probs: Log probability of encoder outputs will be passed to CTC Loss.</dt><dd><p>If joint_ctc_attention is False, return None.</p>
</dd>
</dl>
</li>
</ul>
</p>
</dd>
</dl>
<dl class="py method">
<dt id="kospeech.models.transformer.model.SpeechTransformer.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">inputs</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">input_lengths</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">targets</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em><span class="sig-paren">)</span> → Tuple<span class="p">[</span><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><span class="p">, </span><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><span class="p">, </span><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><span class="p">]</span><a class="reference internal" href="_modules/kospeech/models/transformer/model.html#SpeechTransformer.forward"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.model.SpeechTransformer.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Forward propagate a <cite>inputs</cite> and <cite>targets</cite> pair for training.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>inputs</strong> (<em>torch.FloatTensor</em>) – A input sequence passed to encoder. Typically for inputs this will be a padded
<cite>FloatTensor</cite> of size <code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">seq_length,</span> <span class="pre">dimension)</span></code>.</p></li>
<li><p><strong>input_lengths</strong> (<em>torch.LongTensor</em>) – The length of input tensor. <code class="docutils literal notranslate"><span class="pre">(batch)</span></code></p></li>
<li><p><strong>targets</strong> (<em>torch.LongTensr</em>) – A target sequence passed to decoder. <cite>IntTensor</cite> of size <code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">seq_length)</span></code></p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><p>(Tensor, Tensor, Tensor)</p>
<ul class="simple">
<li><p>predicted_log_probs (torch.FloatTensor): Log probability of model predictions.</p></li>
<li><p>encoder_output_lengths: The length of encoder outputs. <code class="docutils literal notranslate"><span class="pre">(batch)</span></code></p></li>
<li><dl class="simple">
<dt>encoder_log_probs: Log probability of encoder outputs will be passed to CTC Loss.</dt><dd><p>If joint_ctc_attention is False, return None.</p>
</dd>
</dl>
</li>
</ul>
</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
</div>
<div class="section" id="module-kospeech.models.transformer.encoder">
<span id="encoder"></span><h2>Encoder<a class="headerlink" href="#module-kospeech.models.transformer.encoder" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="kospeech.models.transformer.encoder.TransformerEncoder">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.encoder.</code><code class="sig-name descname">TransformerEncoder</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">input_dim</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span></em>, <em class="sig-param"><span class="n">extractor</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.9)">str</a></span> <span class="o">=</span> <span class="default_value">'vgg'</span></em>, <em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">d_ff</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">2048</span></em>, <em class="sig-param"><span class="n">num_layers</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">6</span></em>, <em class="sig-param"><span class="n">num_heads</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">8</span></em>, <em class="sig-param"><span class="n">dropout_p</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)">float</a></span> <span class="o">=</span> <span class="default_value">0.3</span></em>, <em class="sig-param"><span class="n">joint_ctc_attention</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.9)">bool</a></span> <span class="o">=</span> <span class="default_value">False</span></em>, <em class="sig-param"><span class="n">num_classes</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">None</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/encoder.html#TransformerEncoder"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.encoder.TransformerEncoder" title="Permalink to this definition">¶</a></dt>
<dd><p>The TransformerEncoder is composed of a stack of N identical layers.
Each layer has two sub-layers. The first is a multi-head self-attention mechanism,
and the second is a simple, position-wise fully connected feed-forward network.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>input_dim</strong> – dimension of feature vector</p></li>
<li><p><strong>extractor</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.9)"><em>str</em></a>) – convolutional extractor</p></li>
<li><p><strong>d_model</strong> – dimension of model (default: 512)</p></li>
<li><p><strong>d_ff</strong> – dimension of feed forward network (default: 2048)</p></li>
<li><p><strong>num_layers</strong> – number of encoder layers (default: 6)</p></li>
<li><p><strong>num_heads</strong> – number of attention heads (default: 8)</p></li>
<li><p><strong>dropout_p</strong> – probability of dropout (default: 0.3)</p></li>
</ul>
</dd>
</dl>
<dl class="simple">
<dt>Inputs:</dt><dd><ul class="simple">
<li><p><strong>inputs</strong>: list of sequences, whose length is the batch size and within which each sequence is list of tokens</p></li>
<li><p><strong>input_lengths</strong>: list of sequence lengths</p></li>
</ul>
</dd>
</dl>
<dl class="py method">
<dt id="kospeech.models.transformer.encoder.TransformerEncoder.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">inputs</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">input_lengths</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em><span class="sig-paren">)</span> → Tuple<span class="p">[</span><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><span class="p">, </span><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><span class="p">, </span><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><span class="p">]</span><a class="reference internal" href="_modules/kospeech/models/transformer/encoder.html#TransformerEncoder.forward"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.encoder.TransformerEncoder.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Forward propagate a <cite>inputs</cite> for encoder training.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>inputs</strong> (<em>torch.FloatTensor</em>) – A input sequence passed to encoder. Typically for inputs this will be a padded
<cite>FloatTensor</cite> of size <code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">seq_length,</span> <span class="pre">dimension)</span></code>.</p></li>
<li><p><strong>input_lengths</strong> (<em>torch.LongTensor</em>) – The length of input tensor. <code class="docutils literal notranslate"><span class="pre">(batch)</span></code></p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p><ul class="simple">
<li><p>outputs: A output sequence of encoder. <cite>FloatTensor</cite> of size <code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">seq_length,</span> <span class="pre">dimension)</span></code></p></li>
<li><p>output_lengths: The length of encoder outputs. <code class="docutils literal notranslate"><span class="pre">(batch)</span></code></p></li>
<li><dl class="simple">
<dt>encoder_log_probs: Log probability of encoder outputs will be passed to CTC Loss.</dt><dd><p>If joint_ctc_attention is False, return None.</p>
</dd>
</dl>
</li>
</ul>
</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p>(Tensor, Tensor, Tensor)</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="py class">
<dt id="kospeech.models.transformer.encoder.TransformerEncoderLayer">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.encoder.</code><code class="sig-name descname">TransformerEncoderLayer</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">num_heads</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">8</span></em>, <em class="sig-param"><span class="n">d_ff</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">2048</span></em>, <em class="sig-param"><span class="n">dropout_p</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)">float</a></span> <span class="o">=</span> <span class="default_value">0.3</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/encoder.html#TransformerEncoderLayer"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.encoder.TransformerEncoderLayer" title="Permalink to this definition">¶</a></dt>
<dd><p>EncoderLayer is made up of self-attention and feedforward network.
This standard encoder layer is based on the paper “Attention Is All You Need”.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>d_model</strong> – dimension of model (default: 512)</p></li>
<li><p><strong>num_heads</strong> – number of attention heads (default: 8)</p></li>
<li><p><strong>d_ff</strong> – dimension of feed forward network (default: 2048)</p></li>
<li><p><strong>dropout_p</strong> – probability of dropout (default: 0.3)</p></li>
</ul>
</dd>
</dl>
</dd></dl>
</div>
<div class="section" id="module-kospeech.models.transformer.decoder">
<span id="decoder"></span><h2>Decoder<a class="headerlink" href="#module-kospeech.models.transformer.decoder" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="kospeech.models.transformer.decoder.TransformerDecoder">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.decoder.</code><code class="sig-name descname">TransformerDecoder</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">num_classes</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span></em>, <em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">d_ff</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">num_layers</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">6</span></em>, <em class="sig-param"><span class="n">num_heads</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">8</span></em>, <em class="sig-param"><span class="n">dropout_p</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)">float</a></span> <span class="o">=</span> <span class="default_value">0.3</span></em>, <em class="sig-param"><span class="n">pad_id</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">0</span></em>, <em class="sig-param"><span class="n">sos_id</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">1</span></em>, <em class="sig-param"><span class="n">eos_id</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">2</span></em>, <em class="sig-param"><span class="n">max_length</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">400</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/decoder.html#TransformerDecoder"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.decoder.TransformerDecoder" title="Permalink to this definition">¶</a></dt>
<dd><p>The TransformerDecoder is composed of a stack of N identical layers.
Each layer has three sub-layers. The first is a multi-head self-attention mechanism,
and the second is a multi-head attention mechanism, third is a feed-forward network.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>num_classes</strong> – umber of classes</p></li>
<li><p><strong>d_model</strong> – dimension of model</p></li>
<li><p><strong>d_ff</strong> – dimension of feed forward network</p></li>
<li><p><strong>num_layers</strong> – number of decoder layers</p></li>
<li><p><strong>num_heads</strong> – number of attention heads</p></li>
<li><p><strong>dropout_p</strong> – probability of dropout</p></li>
<li><p><strong>pad_id</strong> – identification of pad token</p></li>
<li><p><strong>eos_id</strong> – identification of end of sentence token</p></li>
</ul>
</dd>
</dl>
<dl class="py method">
<dt id="kospeech.models.transformer.decoder.TransformerDecoder.decode">
<code class="sig-name descname">decode</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">encoder_outputs</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">encoder_output_lengths</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em><span class="sig-paren">)</span> → <a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><a class="reference internal" href="_modules/kospeech/models/transformer/decoder.html#TransformerDecoder.decode"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.decoder.TransformerDecoder.decode" title="Permalink to this definition">¶</a></dt>
<dd><p>Decode encoder_outputs.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>encoder_outputs</strong> (<em>torch.FloatTensor</em>) – A output sequence of encoder. <cite>FloatTensor</cite> of size
<code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">seq_length,</span> <span class="pre">dimension)</span></code></p>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>Log probability of model predictions.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><ul class="simple">
<li><p>predicted_log_probs (torch.FloatTensor)</p></li>
</ul>
</p>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt id="kospeech.models.transformer.decoder.TransformerDecoder.forward">
<code class="sig-name descname">forward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">targets</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">encoder_outputs</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">encoder_output_lengths</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em><span class="sig-paren">)</span> → <a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><a class="reference internal" href="_modules/kospeech/models/transformer/decoder.html#TransformerDecoder.forward"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.decoder.TransformerDecoder.forward" title="Permalink to this definition">¶</a></dt>
<dd><p>Forward propagate a <cite>encoder_outputs</cite> for training.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>targets</strong> (<em>torch.LongTensr</em>) – A target sequence passed to decoder. <cite>IntTensor</cite> of size <code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">seq_length)</span></code></p></li>
<li><p><strong>encoder_outputs</strong> (<em>torch.FloatTensor</em>) – A output sequence of encoder. <cite>FloatTensor</cite> of size
<code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">seq_length,</span> <span class="pre">dimension)</span></code></p></li>
<li><p><strong>encoder_output_lengths</strong> – The length of encoder outputs. <code class="docutils literal notranslate"><span class="pre">(batch)</span></code></p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>Log probability of model predictions.</p>
</dd>
<dt class="field-odd">Return type</dt>
<dd class="field-odd"><p><ul class="simple">
<li><p>predicted_log_probs (torch.FloatTensor)</p></li>
</ul>
</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="py class">
<dt id="kospeech.models.transformer.decoder.TransformerDecoderLayer">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.decoder.</code><code class="sig-name descname">TransformerDecoderLayer</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">num_heads</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">8</span></em>, <em class="sig-param"><span class="n">d_ff</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">2048</span></em>, <em class="sig-param"><span class="n">dropout_p</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)">float</a></span> <span class="o">=</span> <span class="default_value">0.3</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/decoder.html#TransformerDecoderLayer"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.decoder.TransformerDecoderLayer" title="Permalink to this definition">¶</a></dt>
<dd><p>DecoderLayer is made up of self-attention, multi-head attention and feedforward network.
This standard decoder layer is based on the paper “Attention Is All You Need”.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>d_model</strong> – dimension of model (default: 512)</p></li>
<li><p><strong>num_heads</strong> – number of attention heads (default: 8)</p></li>
<li><p><strong>d_ff</strong> – dimension of feed forward network (default: 2048)</p></li>
<li><p><strong>dropout_p</strong> – probability of dropout (default: 0.3)</p></li>
</ul>
</dd>
</dl>
</dd></dl>
</div>
<div class="section" id="module-kospeech.models.transformer.sublayers">
<span id="sublayers"></span><h2>Sublayers<a class="headerlink" href="#module-kospeech.models.transformer.sublayers" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="kospeech.models.transformer.sublayers.AddNorm">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.sublayers.</code><code class="sig-name descname">AddNorm</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">sublayer</span><span class="p">:</span> <span class="n">torch.nn.modules.module.Module</span></em>, <em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/sublayers.html#AddNorm"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.sublayers.AddNorm" title="Permalink to this definition">¶</a></dt>
<dd><p>Add & Normalization layer proposed in “Attention Is All You Need”.
Transformer employ a residual connection around each of the two sub-layers,
(Multi-Head Attention & Feed-Forward) followed by layer normalization.</p>
</dd></dl>
<dl class="py class">
<dt id="kospeech.models.transformer.sublayers.PositionwiseFeedForward">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.sublayers.</code><code class="sig-name descname">PositionwiseFeedForward</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">d_ff</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">2048</span></em>, <em class="sig-param"><span class="n">dropout_p</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.9)">float</a></span> <span class="o">=</span> <span class="default_value">0.3</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/sublayers.html#PositionwiseFeedForward"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.sublayers.PositionwiseFeedForward" title="Permalink to this definition">¶</a></dt>
<dd><p>Position-wise Feedforward Networks proposed in “Attention Is All You Need”.
Fully connected feed-forward network, which is applied to each position separately and identically.
This consists of two linear transformations with a ReLU activation in between.
Another way of describing this is as two convolutions with kernel size 1.</p>
</dd></dl>
</div>
<div class="section" id="module-kospeech.models.transformer.embeddings">
<span id="embeddings"></span><h2>Embeddings<a class="headerlink" href="#module-kospeech.models.transformer.embeddings" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="kospeech.models.transformer.embeddings.Embedding">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.embeddings.</code><code class="sig-name descname">Embedding</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">num_embeddings</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span></em>, <em class="sig-param"><span class="n">pad_id</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span></em>, <em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/embeddings.html#Embedding"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.embeddings.Embedding" title="Permalink to this definition">¶</a></dt>
<dd><p>Embedding layer. Similarly to other sequence transduction models, transformer use learned embeddings
to convert the input tokens and output tokens to vectors of dimension d_model.
In the embedding layers, transformer multiply those weights by sqrt(d_model)</p>
</dd></dl>
<dl class="py class">
<dt id="kospeech.models.transformer.embeddings.PositionalEncoding">
<em class="property">class </em><code class="sig-prename descclassname">kospeech.models.transformer.embeddings.</code><code class="sig-name descname">PositionalEncoding</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">d_model</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">512</span></em>, <em class="sig-param"><span class="n">max_len</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.9)">int</a></span> <span class="o">=</span> <span class="default_value">5000</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/embeddings.html#PositionalEncoding"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.embeddings.PositionalEncoding" title="Permalink to this definition">¶</a></dt>
<dd><p>Positional Encoding proposed in “Attention Is All You Need”.
Since transformer contains no recurrence and no convolution, in order for the model to make
use of the order of the sequence, we must add some positional information.</p>
<dl class="simple">
<dt>“Attention Is All You Need” use sine and cosine functions of different frequencies:</dt><dd><p>PE_(pos, 2i) = sin(pos / power(10000, 2i / d_model))
PE_(pos, 2i+1) = cos(pos / power(10000, 2i / d_model))</p>
</dd>
</dl>
</dd></dl>
</div>
<div class="section" id="module-kospeech.models.transformer.mask">
<span id="mask"></span><h2>Mask<a class="headerlink" href="#module-kospeech.models.transformer.mask" title="Permalink to this headline">¶</a></h2>
<dl class="py function">
<dt id="kospeech.models.transformer.mask.get_attn_pad_mask">
<code class="sig-prename descclassname">kospeech.models.transformer.mask.</code><code class="sig-name descname">get_attn_pad_mask</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">inputs</span></em>, <em class="sig-param"><span class="n">input_lengths</span></em>, <em class="sig-param"><span class="n">expand_length</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/mask.html#get_attn_pad_mask"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.mask.get_attn_pad_mask" title="Permalink to this definition">¶</a></dt>
<dd><p>mask position is set to 1</p>
</dd></dl>
<dl class="py function">
<dt id="kospeech.models.transformer.mask.get_decoder_self_attn_mask">
<code class="sig-prename descclassname">kospeech.models.transformer.mask.</code><code class="sig-name descname">get_decoder_self_attn_mask</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">seq_k</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">seq_q</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">pad_id</span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/kospeech/models/transformer/mask.html#get_decoder_self_attn_mask"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.mask.get_decoder_self_attn_mask" title="Permalink to this definition">¶</a></dt>
<dd><p>For masking the decoder self attention</p>
</dd></dl>
<dl class="py function">
<dt id="kospeech.models.transformer.mask.get_non_pad_mask">
<code class="sig-prename descclassname">kospeech.models.transformer.mask.</code><code class="sig-name descname">get_non_pad_mask</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">inputs</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em>, <em class="sig-param"><span class="n">input_lengths</span><span class="p">:</span> <span class="n"><a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a></span></em><span class="sig-paren">)</span> → <a class="reference external" href="https://pytorch.org/docs/master/tensors.html#torch.Tensor" title="(in PyTorch vmaster (1.8.0a0+4b6fea9 ))">torch.Tensor</a><a class="reference internal" href="_modules/kospeech/models/transformer/mask.html#get_non_pad_mask"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#kospeech.models.transformer.mask.get_non_pad_mask" title="Permalink to this definition">¶</a></dt>
<dd><p>Padding position is set to 0, either use input_lengths or pad_id</p>
</dd></dl>
</div>
</div>
</div>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="Jasper.html" class="btn btn-neutral float-right" title="Jasper" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right" aria-hidden="true"></span></a>
<a href="RNN%20Transducer.html" class="btn btn-neutral float-left" title="RNN Transducer" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left" aria-hidden="true"></span> Previous</a>
</div>
<hr/>
<div role="contentinfo">
<p>
© Copyright 2020, Soohwan Kim.
</p>
</div>
Built with <a href="https://www.sphinx-doc.org/">Sphinx</a> using a
<a href="https://github.com/readthedocs/sphinx_rtd_theme">theme</a>
provided by <a href="https://readthedocs.org">Read the Docs</a>.
</footer>
</div>
</div>
</section>
</div>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>