http://index.cslt.org/mediawiki/index.php?action=history&feed=atom&title=2013-12-202013-12-20 - 版本历史2026-04-14T15:24:13Z本wiki的该页面的版本历史MediaWiki 1.23.3http://index.cslt.org/mediawiki/index.php?title=2013-12-20&diff=8812&oldid=prevCslt：以内容“== AM development == === Sparse DNN === * Optimal Brain Damage(OBD). # Online OBD held. # OBD + L1 norm start to investigation. * Efficient computing # Conducti...”创建新页面2013-12-20T01:32:54Z<p>以内容“== AM development == === Sparse DNN === * Optimal Brain Damage(OBD). # Online OBD held. # OBD + L1 norm start to investigation. * Efficient computing # Conducti...”创建新页面</p>
<p><b>新页面</b></p><div>== AM development ==<br />
<br />
=== Sparse DNN ===<br />
<br />
* Optimal Brain Damage(OBD). <br />
<br />
# Online OBD held. <br />
# OBD + L1 norm start to investigation. <br />
<br />
* Efficient computing<br />
<br />
# Conducting rearrangement the matrix structure and compose zero blocks by some smart approaches, leading to better computing speed. <br />
<br />
<br />
=== Efficient DNN training ===<br />
<br />
# Moment-based training. With m=0.2 performs the best on WER. 6.8% improvement on WER. Other settings are tried on 0.05,0.1,0.2,..0.6,0.8,1.0. <br />
# Asymmetric window: left 20, right 5. NN accuracy increase by 7%, however WER is a bit worse than the baseline. Move back to Tencent 100h training. <br />
# Frame-skipping is on implementation. <br />
<br />
=== Optimal phoneset===<br />
<br />
# Experiment 3 phone sets: Tencent, CSLT, PQ<br />
# Some errors occur in pure CHS experiments<br />
<br />
<br />
===Engine optimization===<br />
<br />
* Investigating LOUDS FST. On progress. <br />
<br />
<br />
==LM development==<br />
<br />
===NN LM===<br />
<br />
* Trained with 500M QA data, 110k vocabulary.<br />
* Tested on number of hidden layers (DNN), performance is better for some tests, but not for others. <br />
* Tested on larger projection layer, from 256 to 384, the performance is consistently improved. <br />
<br />
<br />
==Embedded development==<br />
<br />
* Embedded stream mode on progress.<br />
<br />
<br />
==Speech QA==<br />
<br />
* SP-QA accuracy 45.14% in all the input (18*199).<br />
* Investigate the error patterns:<br />
:* 70% errors are caused by incorrect name entity recognition. Working on entity recovery (character, pinyin, ... distance penalty). <br />
:* 8% errors are caused by English names. Use class-based LM to solve the problem. Ready to work.<br />
:* Use N-best to recover errors in QA.</div>Cslt