Blame view
src/doc/examples.dox
18.9 KB
8dcb6dfcb
first commit
|
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 |
// doc/examples.dox
// Copyright 2016 Fred Richardson Allen Guo
// See ../../COPYING for clarification regarding multiple authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
// http://www.apache.org/licenses/LICENSE-2.0
// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
// MERCHANTABLITY OR NON-INFRINGEMENT.
// See the Apache 2 License for the specific language governing permissions and
// limitations under the License.
/**
\page examples Examples included with Kaldi
When you check out the Kaldi source tree (see \ref install), you will find many
sets of example scripts in the egs/ directory. This table summarizes some key
facts about some of those example scripts; however, it it not an exhaustive
list.
<table><tr>
<th>Name</th>
<th>BW</th>
<th>Lang</th>
<th>Train Domain</th>
<th>Train Hours</th>
<th>Train Speakers</th>
<th>License and Availability</th>
<th>Year Released</th>
<th>Speech Style</th>
<th>Test Domain</th>
<th>Kaldi Aprox Perf</th>
<th>Model Type</th>
<th>LM Data</th>
<th>Lexicon</th>
</tr>
<tr>
<!-- Name --> <td>AMI</td>
<!-- BW --> <td>16k</td>
<!-- Lang --> <td>English<br>(+non-native)</td>
<!-- Train Domain --> <td>Microphone: head-mike,<br>single and multiple<br>distance mikes</td>
<!-- Train Hours --> <td>100</td>
<!-- Train Speakers --> <td>123 M<br>66 F</td>
<!-- License/Availability --> <td>Free /<br>Download<br>http://groups.inf.ed.ac.uk/ami/corpus/</td>
<!-- Year Released --> <td>2014</td>
<!-- Speech Style --> <td>Meeting room</td>
<!-- Test Domain --> <td>Same as train<br>no overlap(?)</td>
<!-- Kaldi Aprox Perf --> <td>~25% WER head (T)DNN<br>~45% WER distant (B)LSTM</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>AMI + (opt) Fisher</td>
<!-- Lexicon --> <td>50K (CMU dict +<br>kaldi sources)</td>
</tr>
<tr>
<!-- Name --> <td>Aspire</td>
<!-- BW --> <td></td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Conversational microphone<br>developed on telephone</td>
<!-- Train Hours --> <td>see Fisher</td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td></td>
<!-- Year Released --> <td>2015</td>
<!-- Speech Style --> <td></td>
<!-- Test Domain --> <td></td>
<!-- Kaldi Aprox Perf --> <td>30.8% WER (dev or eval?)</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td></td>
<!-- Lexicon --> <td></td>
</tr>
<tr>
<!-- Name --> <td>WSJ</td>
<!-- BW --> <td>16k</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Clean close-mic<br>read speech</td>
<!-- Train Hours --> <td>80</td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td>LDC<br>LDC93S6B (WSJ0) and LDC94S13B (WSJ1)</td>
<!-- Year Released --> <td>1993</td>
<!-- Speech Style --> <td>Read speech</td>
<!-- Test Domain --> <td>Same</td>
<!-- Kaldi Aprox Perf --> <td>6-7% WER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>same as train</td>
<!-- Lexicon --> <td>20k (CMU dict)</td>
</tr>
<tr>
<!-- Name --> <td>RM</td>
<!-- BW --> <td></td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>read transcript<br>limited vocab and grammar</td>
<!-- Train Hours --> <td></td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td>LDC<br>LDC93S3A</td>
<!-- Year Released --> <td>1987-1989</td>
<!-- Speech Style --> <td>read speech</td>
<!-- Test Domain --> <td>same</td>
<!-- Kaldi Aprox Perf --> <td>1-2% WER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>predefined grammar</td>
<!-- Lexicon --> <td><1K<br>RM dict</td>
</tr>
<tr>
<!-- Name --> <td>Timit</td>
<!-- BW --> <td>16k</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>read transcript<br>very limited grammar</td>
<!-- Train Hours --> <td></td>
<!-- Train Speakers --> <td>630</td>
<!-- License/Availability --> <td></td>
<!-- Year Released --> <td>1986</td>
<!-- Speech Style --> <td>read speech</td>
<!-- Test Domain --> <td>same</td>
<!-- Kaldi Aprox Perf --> <td>~30-40% PER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>none</td>
<!-- Lexicon --> <td>~47 phones</td>
</tr>
<tr>
<!-- Name --> <td>fisher_english</td>
<!-- BW --> <td>8k</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Telephone speech<br>Auto-transcribed<br>(errorful transcriptions)</td>
<!-- Train Hours --> <td>1,600</td>
<!-- Train Speakers --> <td>5203 M<br>7198 F</td>
<!-- License/Availability --> <td>LDC<br>speech: LDC2004S13, LDC2005S13<br>transcript: LDC2004T19, LDC2005T19</td>
<!-- Year Released --> <td>2004/2005</td>
<!-- Speech Style --> <td>CTS</td>
<!-- Test Domain --> <td>Fisher (may<br>overlap witb<br>train)</td>
<!-- Kaldi Aprox Perf --> <td>~22% WER (DNN)</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>LDC Fisher</td>
<!-- Lexicon --> <td>CMU dict<br>Size UNK</td>
</tr>
<tr>
<!-- Name --> <td>Switchboard 1</td>
<!-- BW --> <td>8k</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>CTS</td>
<!-- Train Hours --> <td>300</td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td>LDC<br>Train: LDC97S62<br>Mississippi State transcriptions<br>Eval: LDC2002S09 and LDC2002T43</td>
<!-- Year Released --> <td>1993/1997/2000</td>
<!-- Speech Style --> <td>CTS</td>
<!-- Test Domain --> <td>CTS<br>eval2000 (hub5)</td>
<!-- Kaldi Aprox Perf --> <td>~10% WER (LSTM)</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Mississippi Trans<br>+ (opt) Fisher</td>
<!-- Lexicon --> <td>30K (CMU dict)</td>
</tr>
<tr>
<!-- Name --> <td>Switchboard 1<br>+ Fisher</td>
<!-- BW --> <td>8k</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>CTS</td>
<!-- Train Hours --> <td>see above</td>
<!-- Train Speakers --> <td>see above</td>
<!-- License/Availability --> <td>see above</td>
<!-- Year Released --> <td>see above</td>
<!-- Speech Style --> <td>CTS</td>
<!-- Test Domain --> <td>eval2000<br>rt03</td>
<!-- Kaldi Aprox Perf --> <td>~12% eval2000<br>~19% rt03</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>see above</td>
<!-- Lexicon --> <td>see above</td>
</tr>
<tr>
<!-- Name --> <td>Callhome<br>Egyptian</td>
<!-- BW --> <td></td>
<!-- Lang --> <td>Egyptian<br>Colloquial<br>Arabic</td>
<!-- Train Domain --> <td>CTS</td>
<!-- Train Hours --> <td>120 conv</td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td>LDC<br>Speech : LDC97S45<br>Transcripts : LDC97T19<br>Lexicon : LDC99L22</td>
<!-- Year Released --> <td>1997</td>
<!-- Speech Style --> <td>CTS</td>
<!-- Test Domain --> <td>hub5 arabic<br>LDC2002S22<br>LDC2002T39</td>
<!-- Kaldi Aprox Perf --> <td>50-60% WER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Train trans</td>
<!-- Lexicon --> <td>LDC dict</td>
</tr>
<tr>
<!-- Name --> <td>Corpus of<br>Spontaneous<br>Japanese</td>
<!-- BW --> <td></td>
<!-- Lang --> <td>Japanese</td>
<!-- Train Domain --> <td>Mixed style<br>Close-talking mic</td>
<!-- Train Hours --> <td>650 hours<br>(240 hr train)</td>
<!-- Train Speakers --> <td>>1,400</td>
<!-- License/Availability --> <td>Unclear how to get this<br>http://www.ninjal.ac.jp/english/products/csj/<br>http://pj.ninjal.ac.jp/corpus_center/csj/</td>
<!-- Year Released --> <td>2004</td>
<!-- Speech Style --> <td>Mixed</td>
<!-- Test Domain --> <td></td>
<!-- Kaldi Aprox Perf --> <td>9-10% WER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>UNK</td>
<!-- Lexicon --> <td>UNK</td>
</tr>
<tr>
<!-- Name --> <td>Fisher Spanish<br>Callhome Spanish</td>
<!-- BW --> <td></td>
<!-- Lang --> <td>Caribbean<br>Spanish</td>
<!-- Train Domain --> <td>CTS</td>
<!-- Train Hours --> <td>Fisher: 163 hrs<br>Callhome: 60 hrs?<br>120 30min conv</td>
<!-- Train Speakers --> <td>Fisher: 136<br>Callhome:</td>
<!-- License/Availability --> <td>LDC<br>Fisher speech : LDC96S35<br>Fisher transcripts : LDC96T17<br>Callhome Speech : LDC96S35<br>Callhome Transcripts : LDC96T17</td>
<!-- Year Released --> <td>Fisher: 2010<br>Callhome: 1996</td>
<!-- Speech Style --> <td>CTS</td>
<!-- Test Domain --> <td>Kaldi subset<br>of Fisher</td>
<!-- Kaldi Aprox Perf --> <td>29-30% WER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Fisher trans</td>
<!-- Lexicon --> <td>LDC96L16</td>
</tr>
<tr>
<!-- Name --> <td>Gale Arabic<br>Phase 2</td>
<!-- BW --> <td>16K</td>
<!-- Lang --> <td>Arabic</td>
<!-- Train Domain --> <td>Broadcast<br>Conversational/Report</td>
<!-- Train Hours --> <td>320 train<br>9.3 test</td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td>LDC2013S02 LDC2014S07<br>LDC2013S07 LDC2014T17<br>LDC2013T17<br>LDC2013T04</td>
<!-- Year Released --> <td>Collected<br>2006/2007</td>
<!-- Speech Style --> <td>Broadcast<br>Conversational<br>and Report</td>
<!-- Test Domain --> <td></td>
<!-- Kaldi Aprox Perf --> <td>Report: 13% WER (LSTM)<br>Conver: 28% WER (LSTM)<br>Comb: 24% WER (LSTM)</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>LDC2013T17<br>LDC2013T04<br>LDC2014T17</td>
<!-- Lexicon --> <td>http://alt.qcri.org/</td>
</tr>
<tr>
<!-- Name --> <td>Gale Mandarin</td>
<!-- BW --> <td>16K</td>
<!-- Lang --> <td>Mandarin<br>Chinese</td>
<!-- Train Domain --> <td>Broadcast</td>
<!-- Train Hours --> <td>126</td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td>LDC2013S08 LDC2013T20</td>
<!-- Year Released --> <td>2006-2007</td>
<!-- Speech Style --> <td>Broadcast</td>
<!-- Test Domain --> <td>Same as train</td>
<!-- Kaldi Aprox Perf --> <td>17.5% WER [1]</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>LDC2013S08<br>LDC2013T20</td>
<!-- Lexicon --> <td>Same as HKUST below</td>
</tr>
<tr>
<!-- Name --> <td>hkust<br>EARS RT04F data<br>dev and train [2]</td>
<!-- BW --> <td>8K</td>
<!-- Lang --> <td>Mandarin<br>Chinese</td>
<!-- Train Domain --> <td>Telephone Conversational</td>
<!-- Train Hours --> <td>~145</td>
<!-- Train Speakers --> <td>~873</td>
<!-- License/Availability --> <td>LDC2005S15 LDC2005T32</td>
<!-- Year Released --> <td>2004</td>
<!-- Speech Style --> <td>Conversational</td>
<!-- Test Domain --> <td>Same as train</td>
<!-- Kaldi Aprox Perf --> <td>33.5% CER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Acoustic trans<br>(very little)</td>
<!-- Lexicon --> <td>Both Eng and Man.<br>CMU dict use for Eng<br>mdbg dict use for Man<br>http://www.mdbg.net</td>
</tr>
<tr>
<!-- Name --> <td>librispeech [3]</td>
<!-- BW --> <td>16K</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Read transcription</td>
<!-- Train Hours --> <td>100 - 960<br>(460 </td>
<!-- Train Speakers --> <td>F: 125-1128<br>M: 126-1167</td>
<!-- License/Availability --> <td>http://www.openslr.org/12/</td>
<!-- Year Released --> <td>2015</td>
<!-- Speech Style --> <td>Read trans</td>
<!-- Test Domain --> <td>Librispeech<br></td>
<!-- Kaldi Aprox Perf --> <td>~5% </td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Large (books)</td>
<!-- Lexicon --> <td>cmu (with sequitur)<br>G2P)</td>
</tr>
<tr>
<!-- Name --> <td>reverb</td>
<!-- BW --> <td></td>
<!-- Lang --> <td></td>
<!-- Train Domain --> <td></td>
<!-- Train Hours --> <td></td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td></td>
<!-- Year Released --> <td></td>
<!-- Speech Style --> <td></td>
<!-- Test Domain --> <td></td>
<!-- Kaldi Aprox Perf --> <td></td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td></td>
<!-- Lexicon --> <td></td>
</tr>
<tr>
<!-- Name --> <td>sprakbanken</td>
<!-- BW --> <td></td>
<!-- Lang --> <td>Danish</td>
<!-- Train Domain --> <td>Read transcript?</td>
<!-- Train Hours --> <td>350</td>
<!-- Train Speakers --> <td></td>
<!-- License/Availability --> <td>Free download<br>http://www.nb.no/sprakbanken/#ticketsfrom?lang=en</td>
<!-- Year Released --> <td>2012</td>
<!-- Speech Style --> <td>Read/Dictation</td>
<!-- Test Domain --> <td>Same as train</td>
<!-- Kaldi Aprox Perf --> <td>14% WER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>NST Provided</td>
<!-- Lexicon --> <td>NST Provided?</td>
</tr>
<tr>
<!-- Name --> <td>vystadial_en [4]</td>
<!-- BW --> <td>8Khz</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Telephone, dialog system</td>
<!-- Train Hours --> <td>41</td>
<!-- Train Speakers --> <td>unk</td>
<!-- License/Availability --> <td>Free</td>
<!-- Year Released --> <td>2014</td>
<!-- Speech Style --> <td>Dialog sys</td>
<!-- Test Domain --> <td>Same as train</td>
<!-- Kaldi Aprox Perf --> <td>~11% WER (GMM/HMM)</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Train trans</td>
<!-- Lexicon --> <td>CMU + 250</td>
</tr>
<tr>
<!-- Name --> <td>vystadial_cz [4]</td>
<!-- BW --> <td>8Khz</td>
<!-- Lang --> <td>Czech</td>
<!-- Train Domain --> <td>Telephone, dialog system</td>
<!-- Train Hours --> <td>15</td>
<!-- Train Speakers --> <td>unk</td>
<!-- License/Availability --> <td>Free</td>
<!-- Year Released --> <td>2014</td>
<!-- Speech Style --> <td>Dialog sys</td>
<!-- Test Domain --> <td>Same as train</td>
<!-- Kaldi Aprox Perf --> <td>~50% WER (GMM/HMM)</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Train trans</td>
<!-- Lexicon --> <td>Rule derived</td>
</tr>
<tr>
<!-- Name --> <td>chime3</td>
<!-- BW --> <td>16Khz</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Read trans, simulated<br>and real noise</td>
<!-- Train Hours --> <td>18</td>
<!-- Train Speakers --> <td>WSJ0 + 4</td>
<!-- License/Availability --> <td>Not clear (Chime performers)</td>
<!-- Year Released --> <td>2015</td>
<!-- Speech Style --> <td>Read<br>transcript</td>
<!-- Test Domain --> <td>Same as train<br>(same channels!)</td>
<!-- Kaldi Aprox Perf --> <td>~12% WER real (4 spkrs)<br>~12% WER simu</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Official WSJ0 5K<br>trans</td>
<!-- Lexicon --> <td>WSJ0</td>
</tr>
<tr>
<!-- Name --> <td>voxforge</td>
<!-- BW --> <td>16Khz</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Read trans</td>
<!-- Train Hours --> <td>>75hrs</td>
<!-- Train Speakers --> <td>unk</td>
<!-- License/Availability --> <td>Free GPL</td>
<!-- Year Released --> <td>2008?</td>
<!-- Speech Style --> <td>Read trans</td>
<!-- Test Domain --> <td>unk</td>
<!-- Kaldi Aprox Perf --> <td>unk</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Train</td>
<!-- Lexicon --> <td>cmu + g2p for oov</td>
</tr>
<tr>
<!-- Name --> <td>Tedlium</td>
<!-- BW --> <td>16KHz</td>
<!-- Lang --> <td>English</td>
<!-- Train Domain --> <td>Presentation/talk</td>
<!-- Train Hours --> <td>118</td>
<!-- Train Speakers --> <td>666</td>
<!-- License/Availability --> <td>Free download</td>
<!-- Year Released --> <td>2014?</td>
<!-- Speech Style --> <td>Presentation</td>
<!-- Test Domain --> <td>Same as train</td>
<!-- Kaldi Aprox Perf --> <td>~10% WER</td>
<!-- Model Type --> <td></td>
<!-- LM Data --> <td>Cantab provided LM</td>
<!-- Lexicon --> <td>Cantab provided dict</td>
</tr>
<!-- Add a new row by copying a <tr></tr> block from above
and pasting it above this message. -->
</table>
[1] "Audio Augmentation for Speech Recognition" Tom Ko, Vijayaditya Peddinti, Daniel Povey, Sanjeev Khudanpur.<br>
[2] There should be more Mandarin data from rt04f - 50 hours of dev data I believe (see LDC2004E67, LDC2004E68). There should also be eval data. See https://www.ldc.upenn.edu/collaborations/past-projects/gale/data/gale-pubs.<br>
[3] See http://www.danielpovey.com/files/2015_icassp_librispeech.pdf for details. Acoustic and language models are available online.<br>
[4] See http://www.lrec-conf.org/proceedings/lrec2014/pdf/535_Paper.pdf.
*/
|