Skip to content

P
Paddle

PaddlePaddle / Paddle
大约 1 年 前同步成功

通知 2298
Star 20931
Fork 5422
P
Paddle
切换分支/标签
查找文件 普通视图历史永久链接Permalink
deep_speech_2.html 25.4 KB
Newer Older
T
Deploy to GitHub Pages: ae7d1c1f8c7175d60420dcfefba37c1a2518a536  
Travis CI 已提交
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 


<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
  <meta charset="utf-8">
  
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  
  <title>DeepSpeech2 on PaddlePaddle: Design Doc &mdash; PaddlePaddle  documentation</title>
  

  
  

  

  
  
    

  

  
  
    <link rel="stylesheet" href="../../_static/css/theme.css" type="text/css" />
  

  
  
        <link rel="index" title="Index"
              href="../../genindex.html"/>
        <link rel="search" title="Search" href="../../search.html"/>
    <link rel="top" title="PaddlePaddle  documentation" href="../../index.html"/> 

  <link rel="stylesheet" href="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/css/perfect-scrollbar.min.css" type="text/css" />
  <link rel="stylesheet" href="../../_static/css/override.css" type="text/css" />
  <script>
  var _hmt = _hmt || [];
  (function() {
    var hm = document.createElement("script");
    hm.src = "//hm.baidu.com/hm.js?b9a314ab40d04d805655aab1deee08ba";
    var s = document.getElementsByTagName("script")[0]; 
    s.parentNode.insertBefore(hm, s);
  })();
  </script>

  

  
  <script src="../../_static/js/modernizr.min.js"></script>

</head>

<body class="wy-body-for-nav" role="document">

  
  <header class="site-header">
    <div class="site-logo">
      <a href="/"><img src="../../_static/images/PP_w.png"></a>
    </div>
    <div class="site-nav-links">
      <div class="site-menu">
        <a class="fork-on-github" href="https://github.com/PaddlePaddle/Paddle" target="_blank"><i class="fa fa-github"></i>Fork me on Github</a>
        <div class="language-switcher dropdown">
          <a type="button" data-toggle="dropdown">
            <span>English</span>
            <i class="fa fa-angle-up"></i>
            <i class="fa fa-angle-down"></i>
          </a>
          <ul class="dropdown-menu">
            <li><a href="/doc_cn">中文</a></li>
            <li><a href="/doc">English</a></li>
          </ul>
        </div>
        <ul class="site-page-links">
          <li><a href="/">Home</a></li>
        </ul>
      </div>
      <div class="doc-module">
        
        <ul>
<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_en.html">GET STARTED</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../howto/index_en.html">HOW TO</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../api/index_en.html">API</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mobile/index_en.html">MOBILE</a></li>
</ul>

        
<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>
  </header>
  
  <div class="main-content-wrap">

    
    <nav class="doc-menu-vertical" role="navigation">
        
          
          <ul>
<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_en.html">GET STARTED</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../getstarted/build_and_install/index_en.html">Install and Build</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/pip_install_en.html">Install Using pip</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/docker_install_en.html">Run in Docker Containers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/dev/build_en.html">Build using Docker</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/build_from_source_en.html">Build from Sources</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../howto/index_en.html">HOW TO</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cmd_parameter/index_en.html">Set Command-line Parameters</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/use_case_en.html">Use Case</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/arguments_en.html">Argument Outline</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/detail_introduction_en.html">Detail Description</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cluster/cluster_train_en.html">Distributed Training</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cluster/fabric_en.html">fabric</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cluster/openmpi_en.html">openmpi</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cluster/k8s_en.html">kubernetes</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cluster/k8s_aws_en.html">kubernetes on AWS</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/new_layer_en.html">Write New Layers</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/contribute_to_paddle_en.html">Contribute Code</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/write_docs_en.html">Contribute Documentation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/deep_model/rnn/index_en.html">RNN Models</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/rnn_config_en.html">RNN Configuration</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/optimization/gpu_profiling_en.html">Tune GPU Performance</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../api/index_en.html">API</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/model_configs.html">Model Configuration</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/activation.html">Activation</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/layer.html">Layers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/evaluators.html">Evaluators</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/optimizer.html">Optimizer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/pooling.html">Pooling</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/networks.html">Networks</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/attr.html">Parameter Attribute</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/data.html">Data Reader Interface and DataSets</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/data/data_reader.html">Data Reader Interface</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/data/image.html">Image Interface</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/data/dataset.html">Dataset</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/run_logic.html">Training and Inference</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/fluid.html">Fluid</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/layers.html">Layers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/data_feeder.html">DataFeeder</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/executor.html">Executor</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/initializer.html">Initializer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/evaluator.html">Evaluator</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/nets.html">Nets</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/optimizer.html">Optimizer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/param_attr.html">ParamAttr</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/profiler.html">Profiler</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/regularizer.html">Regularizer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/fluid/io.html">IO</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../mobile/index_en.html">MOBILE</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_android_en.html">Build PaddlePaddle for Android</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_ios_en.html">Build PaddlePaddle for iOS</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_raspberry_en.html">Build PaddlePaddle for Raspberry Pi</a></li>
</ul>
</li>
</ul>

        
    </nav>
    
    <section class="doc-content-wrap">

      

 







<div role="navigation" aria-label="breadcrumbs navigation">
  <ul class="wy-breadcrumbs">
      
    <li>DeepSpeech2 on PaddlePaddle: Design Doc</li>
  </ul>
</div>
      
      <div class="wy-nav-content" id="doc-content">
        <div class="rst-content">
          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
           <div itemprop="articleBody">
            
  <div class="section" id="deepspeech2-on-paddlepaddle-design-doc">
<span id="deepspeech2-on-paddlepaddle-design-doc"></span><h1>DeepSpeech2 on PaddlePaddle: Design Doc<a class="headerlink" href="#deepspeech2-on-paddlepaddle-design-doc" title="Permalink to this headline"></a></h1>
<p>We are planning to build Deep Speech 2 (DS2) [<a class="reference external" href="#references">1</a>], a powerful Automatic Speech Recognition (ASR) engine,  on PaddlePaddle. For the first-stage plan, we have the following short-term goals:</p>
<ul class="simple">
<li>Release a basic distributed implementation of DS2 on PaddlePaddle.</li>
<li>Contribute a chapter of Deep Speech to PaddlePaddle Book.</li>
</ul>
<p>Intensive system optimization and low-latency inference library (details in [<a class="reference external" href="#references">1</a>]) are not yet covered in this first-stage plan.</p>
<div class="section" id="table-of-contents">
<span id="table-of-contents"></span><h2>Table of Contents<a class="headerlink" href="#table-of-contents" title="Permalink to this headline"></a></h2>
<ul class="simple">
<li><a class="reference external" href="#tasks">Tasks</a></li>
<li><a class="reference external" href="#task-dependency">Task Dependency</a></li>
<li><a class="reference external" href="#design-details">Design Details</a><ul>
<li><a class="reference external" href="#overview">Overview</a></li>
<li><a class="reference external" href="#row-convolution">Row Convolution</a></li>
<li><a class="reference external" href="#beam-search-with-ctc-and-lm">Beam Search With CTC and LM</a></li>
</ul>
</li>
<li><a class="reference external" href="#future-work">Future Work</a></li>
<li><a class="reference external" href="#references">References</a></li>
</ul>
</div>
<div class="section" id="tasks">
<span id="tasks"></span><h2>Tasks<a class="headerlink" href="#tasks" title="Permalink to this headline"></a></h2>
<p>We roughly break down the project into 14 tasks:</p>
<ol class="simple">
<li>Develop an <strong>audio data provider</strong>:<ul>
<li>Json filelist generator.</li>
<li>Audio file format transformer.</li>
<li>Spectrogram feature extraction, power normalization etc.</li>
<li>Batch data reader with SortaGrad.</li>
<li>Data augmentation (optional).</li>
<li>Prepare (one or more) public English data sets &amp; baseline.</li>
</ul>
</li>
<li>Create a <strong>simplified DS2 model configuration</strong>:<ul>
<li>With only fixed-length (by padding) audio sequences (otherwise need <em>Task 3</em>).</li>
<li>With only bidirectional-GRU (otherwise need <em>Task 4</em>).</li>
<li>With only greedy decoder (otherwise need <em>Task 5, 6</em>).</li>
</ul>
</li>
<li>Develop to support <strong>variable-shaped</strong> dense-vector (image) batches of input data.<ul>
<li>Update <code class="docutils literal"><span class="pre">DenseScanner</span></code> in <code class="docutils literal"><span class="pre">dataprovider_converter.py</span></code>, etc.</li>
</ul>
</li>
<li>Develop a new <strong>lookahead-row-convolution layer</strong> (See [<a class="reference external" href="#references">1</a>] for details):<ul>
<li>Lookahead convolution windows.</li>
<li>Within-row convolution, without kernels shared across rows.</li>
</ul>
</li>
<li>Build KenLM <strong>language model</strong> (5-gram) for beam search decoder:<ul>
<li>Use KenLM toolkit.</li>
<li>Prepare the corpus &amp; train the model.</li>
<li>Create infererence interfaces (for Task 6).</li>
</ul>
</li>
<li>Develop a <strong>beam search decoder</strong> with CTC + LM + WORDCOUNT:<ul>
<li>Beam search with CTC.</li>
<li>Beam search with external custom scorer (e.g. LM).</li>
<li>Try to design a more general beam search interface.</li>
</ul>
</li>
<li>Develop a <strong>Word Error Rate evaluator</strong>:<ul>
<li>update <code class="docutils literal"><span class="pre">ctc_error_evaluator</span></code>(CER) to support WER.</li>
</ul>
</li>
<li>Prepare internal dataset for Mandarin (optional):<ul>
<li>Dataset, baseline, evaluation details.</li>
<li>Particular data preprocessing for Mandarin.</li>
<li>Might need cooperating with the Speech Department.</li>
</ul>
</li>
<li>Create <strong>standard DS2 model configuration</strong>:<ul>
<li>With variable-length audio sequences (need <em>Task 3</em>).</li>
<li>With unidirectional-GRU + row-convolution (need <em>Task 4</em>).</li>
<li>With CTC-LM beam search decoder (need <em>Task 5, 6</em>).</li>
</ul>
</li>
<li>Make it run perfectly on <strong>clusters</strong>.</li>
<li>Experiments and <strong>benchmarking</strong> (for accuracy, not efficiency):<ul>
<li>With public English dataset.</li>
<li>With internal (Baidu) Mandarin dataset (optional).</li>
</ul>
</li>
<li>Time <strong>profiling</strong> and optimization.</li>
<li>Prepare <strong>docs</strong>.</li>
<li>Prepare PaddlePaddle <strong>Book</strong> chapter with a simplified version.</li>
</ol>
</div>
<div class="section" id="task-dependency">
<span id="task-dependency"></span><h2>Task Dependency<a class="headerlink" href="#task-dependency" title="Permalink to this headline"></a></h2>
<p>Tasks parallelizable within phases:</p>
<p>Roadmap     | Description                               | Parallelizable Tasks
&#8212;&#8212;&#8212;&#8211; | :&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;     | :&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8211;
Phase I     | Simplified model &amp; components             | <em>Task 1</em> ~ <em>Task 8</em>
Phase II    | Standard model &amp; benchmarking &amp; profiling | <em>Task 9</em> ~ <em>Task 12</em>
Phase III   | Documentations                            | <em>Task13</em> ~ <em>Task14</em></p>
<p>Issue for each task will be created later. Contributions, discussions and comments are all highly appreciated and welcomed!</p>
</div>
<div class="section" id="design-details">
<span id="design-details"></span><h2>Design Details<a class="headerlink" href="#design-details" title="Permalink to this headline"></a></h2>
<div class="section" id="overview">
<span id="overview"></span><h3>Overview<a class="headerlink" href="#overview" title="Permalink to this headline"></a></h3>
<p>Traditional <strong>ASR</strong> (Automatic Speech Recognition) pipelines require great human efforts devoted to elaborately tuning multiple hand-engineered components (e.g. audio feature design, accoustic model, pronuncation model and language model etc.). <strong>Deep Speech 2</strong> (<strong>DS2</strong>) [<a class="reference external" href="#references">1</a>], however, trains such ASR models in an end-to-end manner, replacing most intermediate modules with only a single deep network architecture. With scaling up both the data and model sizes, DS2 achieves a very significant performance boost.</p>
<p>Please read Deep Speech 2 [<a class="reference external" href="#references">1</a>,<a class="reference external" href="#references">2</a>] paper for more background knowledge.</p>
<p>The classical DS2 network contains 15 layers (from bottom to top):</p>
<ul class="simple">
<li><strong>Two</strong> data layers (audio spectrogram, transcription text)</li>
<li><strong>Three</strong> 2D convolution layers</li>
<li><strong>Seven</strong> uni-directional simple-RNN layers</li>
<li><strong>One</strong> lookahead row convolution layers</li>
<li><strong>One</strong> fully-connected layers</li>
<li><strong>One</strong> CTC-loss layer</li>
</ul>
<div align="center">
<img src="image/ds2_network.png" width=350><br/>
Figure 1. Archetecture of Deep Speech 2 Network.
</div><p>We don&#8217;t have to persist on this 2-3-7-1-1-1 depth [<a class="reference external" href="#references">2</a>]. Similar networks with different depths might also work well. As in [<a class="reference external" href="#references">1</a>], authors use a different depth (e.g. 2-2-3-1-1-1) for final experiments.</p>
<p>Key ingredients about the layers:</p>
<ul class="simple">
<li><strong>Data Layers</strong>:<ul>
<li>Frame sequences data of audio <strong>spectrogram</strong> (with FFT).</li>
<li>Token sequences data of <strong>transcription</strong> text (labels).</li>
<li>These two type of sequences do not have the same lengthes, thus a CTC-loss layer is required.</li>
</ul>
</li>
<li><strong>2D Convolution Layers</strong>:<ul>
<li>Not only temporal convolution, but also <strong>frequency convolution</strong>. Like a 2D image convolution, but with a variable dimension (i.e. temporal dimension).</li>
<li>With striding for only the first convlution layer.</li>
<li>No pooling for all convolution layers.</li>
</ul>
</li>
<li><strong>Uni-directional RNNs</strong><ul>
<li>Uni-directional + row convolution: for low-latency inference.</li>
<li>Bi-direcitional + without row convolution: if we don&#8217;t care about the inference latency.</li>
</ul>
</li>
<li><strong>Row convolution</strong>:<ul>
<li>For looking only a few steps ahead into the feature, instead of looking into a whole sequence in bi-directional RNNs.</li>
<li>Not nessesary if with bi-direcitional RNNs.</li>
<li>&#8220;<strong>Row</strong>&#8221; means convolutions are done within each frequency dimension (row), and no convolution kernels shared across.</li>
</ul>
</li>
<li><strong>Batch Normalization Layers</strong>:<ul>
<li>Added to all above layers (except for data and loss layer).</li>
<li>Sequence-wise normalization for RNNs: BatchNorm only performed on input-state projection and not state-state projection, for efficiency consideration.</li>
</ul>
</li>
</ul>
<p>Required Components                     | PaddlePaddle Support                      | Need to Develop
:&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;-  | :&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8211;   | :&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8211;
Data Layer I (Spectrogram)              | Not supported yet.                        |  TBD (Task 3)
Data Layer II (Transcription)           | <code class="docutils literal"><span class="pre">paddle.data_type.integer_value_sequence</span></code> | -
2D Convolution Layer                    | <code class="docutils literal"><span class="pre">paddle.layer.image_conv_layer</span></code>           | -
DataType Converter (vec2seq)            | <code class="docutils literal"><span class="pre">paddle.layer.block_expand</span></code>               | -
Bi-/Uni-directional RNNs                | <code class="docutils literal"><span class="pre">paddle.layer.recurrent_group</span></code>            | -
Row Convolution Layer                   | Not supported yet.                        | TBD (Task 4)
CTC-loss Layer                          | <code class="docutils literal"><span class="pre">paddle.layer.warp_ctc</span></code>                   | -
Batch Normalization Layer               | <code class="docutils literal"><span class="pre">paddle.layer.batch_norm</span></code>                 | -
CTC-Beam search                         | Not supported yet.                        | TBD (Task 6)</p>
</div>
<div class="section" id="row-convolution">
<span id="row-convolution"></span><h3>Row Convolution<a class="headerlink" href="#row-convolution" title="Permalink to this headline"></a></h3>
<p>TODO by Assignees</p>
</div>
<div class="section" id="beam-search-with-ctc-and-lm">
<span id="beam-search-with-ctc-and-lm"></span><h3>Beam Search with CTC and LM<a class="headerlink" href="#beam-search-with-ctc-and-lm" title="Permalink to this headline"></a></h3>
<div align="center">
<img src="image/beam_search.png" width=600><br/>
Figure 2. Algorithm for CTC Beam Search Decoder.
</div><ul class="simple">
<li>The <strong>Beam Search Decoder</strong> for DS2 CTC-trained network follows the similar approach in [<a class="reference external" href="#references">3</a>] as shown in Figure 2, with two important modifications for the ambiguous parts:<ul>
<li><ol class="first">
<li>in the iterative computation of probabilities, the assignment operation is changed to accumulation for one prefix may comes from different paths;</li>
</ol>
</li>
<li><ol class="first">
<li>the if condition <code class="docutils literal"><span class="pre">if</span> <span class="pre">l^+</span> <span class="pre">not</span> <span class="pre">in</span> <span class="pre">A_prev</span> <span class="pre">then</span></code> after probabilities&#8217; computation is deprecated for it is hard to understand and seems unnecessary.</li>
</ol>
</li>
</ul>
</li>
<li>An <strong>external scorer</strong> would be passed into the decoder to evaluate a candidate prefix during decoding whenever a white space appended in English decoding and any character appended in Mandarin decoding.</li>
<li>Such external scorer consists of language model, word count or any other custom scorers.</li>
<li>The <strong>language model</strong> is built from Task 5, with parameters should be carefully tuned to achieve minimum WER/CER (c.f. Task 7)</li>
<li>This decoder needs to perform with <strong>high efficiency</strong> for the convenience of parameters tuning and speech recognition in reality.</li>
</ul>
</div>
</div>
<div class="section" id="future-work">
<span id="future-work"></span><h2>Future Work<a class="headerlink" href="#future-work" title="Permalink to this headline"></a></h2>
<ul class="simple">
<li>Efficiency Improvement</li>
<li>Accuracy Improvement</li>
<li>Low-latency Inference Library</li>
<li>Large-scale benchmarking</li>
</ul>
</div>
<div class="section" id="references">
<span id="references"></span><h2>References<a class="headerlink" href="#references" title="Permalink to this headline"></a></h2>
<ol class="simple">
<li>Dario Amodei, etc., <a class="reference external" href="http://proceedings.mlr.press/v48/amodei16.pdf">Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin</a>. ICML 2016.</li>
<li>Dario Amodei, etc., <a class="reference external" href="https://arxiv.org/abs/1512.02595">Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin</a>.   arXiv:1512.02595.</li>
<li>Awni Y. Hannun, etc. <a class="reference external" href="https://arxiv.org/abs/1408.2873">First-Pass Large Vocabulary Continuous Speech Recognition using Bi-Directional Recurrent DNNs</a>. arXiv:1408.2873</li>
</ol>
</div>
</div>


           </div>
          </div>
          <footer>
  

  <hr/>

  <div role="contentinfo">
    <p>
        &copy; Copyright 2016, PaddlePaddle developers.

    </p>
  </div>
  Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/snide/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">
        var DOCUMENTATION_OPTIONS = {
            URL_ROOT:'../../',
            VERSION:'',
            COLLAPSE_INDEX:false,
            FILE_SUFFIX:'.html',
            HAS_SOURCE:  true,
            SOURCELINK_SUFFIX: ".txt",
        };
    </script>
      <script type="text/javascript" src="../../_static/jquery.js"></script>
      <script type="text/javascript" src="../../_static/underscore.js"></script>
      <script type="text/javascript" src="../../_static/doctools.js"></script>
      <script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
       
  

  
  
    <script type="text/javascript" src="../../_static/js/theme.js"></script>
  
  
  <script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.7/js/bootstrap.min.js" integrity="sha384-Tc5IQib027qvyjSMfHjOMaLkfuWVxZxUPnCJA7l2mCWNIpG9mGCD8wGNIcPD7Txa" crossorigin="anonymous"></script>
  <script src="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/js/perfect-scrollbar.jquery.min.js"></script>
  <script src="../../_static/js/paddle_doc_init.js"></script> 

</body>
</html>