http://index.cslt.org/mediawiki/index.php?action=history&feed=atom&title=Sinovoice-2016-5-19
Sinovoice-2016-5-19 - 版本历史
2026-04-11T10:27:19Z
本wiki的该页面的版本历史
MediaWiki 1.23.3
http://index.cslt.org/mediawiki/index.php?title=Sinovoice-2016-5-19&diff=20350&oldid=prev
2016年5月19日 (四) 05:29 Zhangzy
2016-05-19T05:29:59Z
<p></p>
<table class='diff diff-contentalign-left'>
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr style='vertical-align: top;'>
<td colspan='2' style="background-color: white; color:black; text-align: center;">←上一版本</td>
<td colspan='2' style="background-color: white; color:black; text-align: center;">2016年5月19日 (四) 05:29的版本</td>
</tr><tr><td colspan="2" class="diff-lineno">第11行：</td>
<td colspan="2" class="diff-lineno">第11行：</td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* WeiYu</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* WeiYu</div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">:* 8k more data</ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>:* 50h for training</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>:* 50h for training</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>:* 120h labeled ready</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>:* 120h labeled ready</div></td></tr>
</table>
Zhangzy
http://index.cslt.org/mediawiki/index.php?title=Sinovoice-2016-5-19&diff=20349&oldid=prev
Zhaomy：/* Embedding */
2016-05-19T05:01:26Z
<p><span dir="auto"><span class="autocomment">Embedding</span></span></p>
<table class='diff diff-contentalign-left'>
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr style='vertical-align: top;'>
<td colspan='2' style="background-color: white; color:black; text-align: center;">←上一版本</td>
<td colspan='2' style="background-color: white; color:black; text-align: center;">2016年5月19日 (四) 05:01的版本</td>
</tr><tr><td colspan="2" class="diff-lineno">第109行：</td>
<td colspan="2" class="diff-lineno">第109行：</td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>===Embedding===</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>===Embedding===</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* The size of nnet1 AM is 6.4M (3M after decomposition). So we need to control AM size within 10M.</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* The size of nnet1 AM is 6.4M (3M after decomposition). So we need to control AM size within 10M.</div></td></tr>
<tr><td class='diff-marker'>−</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">* 5*576-2400 TDNN model training done. AM size is about 17M</del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* 5*500-2400 TDNN model on training.</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* 5*500-2400 TDNN model on training.</div></td></tr>
<tr><td class='diff-marker'>−</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div>* MPE training <del class="diffchange diffchange-inline">2/4 part</del></div></td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div>* <ins class="diffchange diffchange-inline">no-svd model, </ins>MPE training <ins class="diffchange diffchange-inline">done</ins></div></td></tr>
<tr><td class='diff-marker'>−</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del class="diffchange diffchange-inline">:</del>* <del class="diffchange diffchange-inline">test8000:wer16</del>, <del class="diffchange diffchange-inline">test10000:wer30</del></div></td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div>* <ins class="diffchange diffchange-inline">svd-100 model</ins>, <ins class="diffchange diffchange-inline">MPE training 2/4 epochs finished.</ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>  LM=1e-5, beam=9, max-active=5000</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>  LM=1e-5, beam=9, max-active=5000</div></td></tr>
</table>
Zhaomy
http://index.cslt.org/mediawiki/index.php?title=Sinovoice-2016-5-19&diff=20348&oldid=prev
Zhaomy：/* Embedding */
2016-05-19T04:57:04Z
<p><span dir="auto"><span class="autocomment">Embedding</span></span></p>
<table class='diff diff-contentalign-left'>
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr style='vertical-align: top;'>
<td colspan='2' style="background-color: white; color:black; text-align: center;">←上一版本</td>
<td colspan='2' style="background-color: white; color:black; text-align: center;">2016年5月19日 (四) 04:57的版本</td>
</tr><tr><td colspan="2" class="diff-lineno">第113行：</td>
<td colspan="2" class="diff-lineno">第113行：</td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* MPE training 2/4 part</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>* MPE training 2/4 part</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>:* test8000:wer16, test10000:wer30</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>:* test8000:wer16, test10000:wer30</div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> LM=1e-5, beam=9, max-active=5000</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  =============================================================================================================</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  |        AM / testset              |  test_1000ju  |  test_2000ju  |  test_8000ju  |  test_10000ju  |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  -------------------------------------------------------------------------------------------------------------</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet1 4*600+800 xEnt (6.4M)      |    25.30    |    40.48    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet1 4*600+800 mpe  (6.4M)      |    20.75    |    35.33    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  -------------------------------------------------------------------------------------------------------------</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet3 5*500 mpe (13M)            |    16.18    |    29.53    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet3 5*500 svd-100 mpe (9.5M)    |    17.87    |    30.39    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  =============================================================================================================</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> LM=1e-6, beam=9, max-active=5000</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  =============================================================================================================</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  |        AM / testset              |  test_1000ju  |  test_2000ju  |  test_8000ju  |  test_10000ju  |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  -------------------------------------------------------------------------------------------------------------</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet1 4*600+800 xEnt (6.4M)      |    21.09    |    36.23    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet1 4*600+800 mpe  (6.4M)      |    16.44    |    30.75    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  -------------------------------------------------------------------------------------------------------------</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet3 5*500 mpe (13M)            |    13.39    |    25.90    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  | nnet3 5*500 svd-100 mpe (9.5M)    |    14.49    |    26.55    |              |                |</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color:black; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">  =============================================================================================================</ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>===Character LM===</div></td><td class='diff-marker'> </td><td style="background-color: #f9f9f9; color: #333333; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #e6e6e6; vertical-align: top; white-space: pre-wrap;"><div>===Character LM===</div></td></tr>
</table>
Zhaomy
http://index.cslt.org/mediawiki/index.php?title=Sinovoice-2016-5-19&diff=20347&oldid=prev
Zhaomy：以“==Data== *16K LingYun :* 2000h data ready :* 4300h real-env data to label * YueYu :* Total 250h(190h-YueYu + 60h-English) :* Add 60h YueYu :* CER: 75%->76% * WeiY...”为内容创建页面
2016-05-19T04:45:27Z
<p>以“==Data== *16K LingYun :* 2000h data ready :* 4300h real-env data to label * YueYu :* Total 250h(190h-YueYu + 60h-English) :* Add 60h YueYu :* CER: 75%->76% * WeiY...”为内容创建页面</p>
<p><b>新页面</b></p><div>==Data==<br />
<br />
*16K LingYun<br />
:* 2000h data ready<br />
:* 4300h real-env data to label<br />
<br />
* YueYu<br />
:* Total 250h(190h-YueYu + 60h-English)<br />
:* Add 60h YueYu<br />
:* CER: 75%->76%<br />
<br />
* WeiYu<br />
:* 50h for training<br />
:* 120h labeled ready<br />
<br />
* PingAn<br />
:*100h User data done<br />
<br />
==Model training==<br />
==Deletion Error Promblem==<br />
* Add one noise phone to alleviate the silence over-training<br />
* Omit sil accuracy in discriminative training<br />
* H smoothing of XEnt and MPE<br />
* Add one silence arc from start-state to end-state<br />
<br />
===Big-Model Training===<br />
====16k====<br />
<br />
====8k=====<br />
=====Model=====<br />
* Add noise phone<br />
:* 1300.mdl done<br />
<br />
* CNN + TDNN<br />
:* 280/900 mdl<br />
:* Need about 12 days Xent<br />
<br />
=====Project=====<br />
* PingAn<br />
:* Add noise phone to phone-list<br />
<br />
=========================================================================================<br />
| AM / config | all | KeHu wer || KeHu no-ins |<br />
-----------------------------------------------------------------------------------------<br />
| tdnn 7-2048 xEnt | 16.45 | 36.49 || 25.18 |<br />
| tdnn 7-2048 MPE | 15.22 | 32.77 || 23.48 |<br />
| tdnn 7-2048 MPE adapt-PABX | 14.67 | 31.33 || 22.76 |<br />
-----------------------------------------------------------------------------------------<br />
| tdnn 7-1024 xEnt | 16.60 | 35.91 || 25.58 |<br />
| tdnn 7-1024 MPE 2e-6 | 15.67 | 32.77 || 26.09 |<br />
| tdnn 7-1024 MPE 2e-5 1.mdl | 15.54 | 32.77 || 26.29 |<br />
| tdnn 7-1024 MPE 1e-5 4.mdl | 15.76 | 33.55 || 27.20 |<br />
| tdnn 7-1024 MPE adapt-PABX | 14.80 | 30.48 || 22.56 |<br />
-----------------------------------------------------------------------------------------<br />
| spn 7-1024 xEnt | 16.49 | 36.23 || 24.59 |<br />
| spn 7-1024 xEnt xEnt-PA_user 101.mdl| 16.19 | 33.22 || 22.69 |<br />
| spn 7-1024 xEnt xEnt-PA_user mpe | 15.24 | 32.77 || 21.65 |<br />
| spn 7-1024 MPE-1000H 23.mdl | 15.29 | 33.09 || 21.65 |<br />
| spn 7-1024 MPE adapt-PA_all 29.mdl | 15.11 | 33.42 || 21.84 |<br />
| spn 7-1024 MPE adapt-PA_user 2e-5 | 15.31 | 31.79 || 20.14 |<br />
| spn 7-1024 MPE adapt-PA_user Hs 2e-5| 15.32 | 32.24 || 20.93 |<br />
=========================================================================================<br />
<br />
Error detail:<br />
============================================================================================<br />
| AM / error | tot_err | ins | del | sub | wer |wer-no-ins|<br />
--------------------------------------------------------------------------------------------<br />
| tdnn 7-1024 xEnt | 549 | 158 | 75 | 316 | 35.91 | 25.58 |<br />
| tdnn 7-1024 MPE | 501 | 102 | 140 | 259 | 32.77 | 26.09 |<br />
| tdnn 7-1024 MPE adapt PABX | 477 | 132 | 92 | 253 | 31.20 | 22.56 |<br />
--------------------------------------------------------------------------------------------<br />
| spn 7-1024 xEnt | 554 | 178 | 66 | 310 | 36.23 | 24.59 |<br />
| spn 7-1024 MPE 1000H | 506 | 175 | 41 | 290 | 33.09 | 21.65 |<br />
| spn 7-1024 MPE adapt User | 486 | 178 | 43 | 265 | 31.79 | 20.14 |<br />
============================================================================================<br />
<br />
* LiaoNingYiDong:<br />
<br />
=================================================================<br />
| AM / config | LNYD |<br />
-----------------------------------------------------------------<br />
| tdnn 7-2048 xEnt | 21.51 |<br />
| tdnn 7-2048 MPE | 20.09 |<br />
| tdnn 7-2048 MPE adapt-LNYD | 17.92 |<br />
-----------------------------------------------------------------<br />
| tdnn 7-1024 xEnt | 21.72 |<br />
| tdnn 7-1024 MPE | 20.99 |<br />
| tdnn 7-1024 MPE adapt-LNYD | |<br />
| cnn 7-1024 xEnt 500.mdl | 21.02 |<br />
-----------------------------------------------------------------<br />
| spn 7-1024 xEnt | 21.70 |<br />
| spn 7-1024 MPE-1000H 23.mdl | 19.97 |<br />
| spn 7-1024 MPE adapt-LNYD | 18.67 |<br />
| spn cnn 7-1024 xEnt 300.mdl | 22.26 |<br />
=================================================================<br />
<br />
Error detail:<br />
==================================================================================<br />
| AM / error | tot_err | ins | del | sub | wer |<br />
----------------------------------------------------------------------------------<br />
| tdnn 7-1024 xEnt | 5873 | 879 | 1364 | 3630 | 21.72 |<br />
| tdnn 7-1024 MPE | 5675 | 710 | 1839 | 3126 | 20.99 |<br />
----------------------------------------------------------------------------------<br />
| spn 7-1024 xEnt | 5866 | 955 | 1353 | 3558 | 21.70 |<br />
| spn 7-1024 MPE 1000H | 5398 | 992 | 937 | 3469 | 19.97 |<br />
==================================================================================<br />
<br />
<br />
===Embedding===<br />
* The size of nnet1 AM is 6.4M (3M after decomposition). So we need to control AM size within 10M.<br />
* 5*576-2400 TDNN model training done. AM size is about 17M<br />
* 5*500-2400 TDNN model on training.<br />
* MPE training 2/4 part<br />
:* test8000:wer16, test10000:wer30<br />
<br />
===Character LM===<br />
* Except Sogou-2T, 9-gram has been done.<br />
* Add word boundary tag to Character-LM trainig done<br />
:* 9-gram<br />
:* Except Weibo & Sogou-2T<br />
:* 1e-7(13M) wer17.91 compared with 1e-7(no-boundary,71M) 13.4<br />
:* 1e-8(54M) wer17.54<br />
<br />
* Prepare specific domain vocabulary<br />
:* Dianxin/Baoxian/Dianli<br />
<br />
*DT lm training<br />
:* ReFr<br />
<br />
*Merge Character-LM & word-LM<br />
:* Union<br />
:* Compose, success.<br />
* 2-step decoding: first, character-based LM. Then, word-based LM.<br />
<br />
===SiaSong Robot===<br />
* Beam-forming algorithm test<br />
* NN-model based beam-forming<br />
<br />
===Project===<br />
* Pingan & Yueyu Deletion error too more<br />
:* TDNN deletion error rate > DNN deletion error rate<br />
:* TDNN Silence scale is too sensitive for different test cases.<br />
<br />
==SID==<br />
===Digit===<br />
* Engine Package</div>
Zhaomy