http://index.cslt.org/mediawiki/index.php?action=history&feed=atom&title=2014-08-29 2026-04-16T20:39:33Z 本wiki的该页面的版本历史 MediaWiki 1.23.3 http://index.cslt.org/mediawiki/index.php?title=2014-08-29&diff=10824&oldid=prev 2014-08-29T02:14:01Z

<p>以内容“==Resoruce Building== == Leftover questions== * Investigating LOUDS FST. * CLG embedded decoder plus online compiler. * DNN-GMM co-training * NN LM == AM developmen...”创建新页面</p> <p><b>新页面</b></p><div>==Resoruce Building==<br /> <br /> == Leftover questions==<br /> <br /> * Investigating LOUDS FST. <br /> * CLG embedded decoder plus online compiler.<br /> * DNN-GMM co-training<br /> * NN LM<br /> <br /> == AM development ==<br /> <br /> === Sparse DNN ===<br /> * WJS sparse DNN does not obtain further improvement<br /> <br /> ===Noise training===<br /> <br /> :* Error found in data setting. Re-run the test with gamma=20,30<br /> :* Re-run test with gamma=1,0.1<br /> :* Noisy training journal paper almost done.<br /> <br /> ==Drop out &amp; Rectification &amp; convolutive network==<br /> <br /> * Change learning to 0.001, the training process can be started: <br /> *# change the drop probability from 0.5 to 0.2. Frame accuracy is improved. WER seems problematic.<br /> *# Experiment learning rate 1 and 8, NA <br /> <br /> * Rectification<br /> # Rectification itself failed with large weights.<br /> # Including L1 penalty enables the training but got very bad performance.<br /> # Try to set the maximum value with rectifier<br /> <br /> * Convolutive network<br /> # Test more configurations <br /> <br /> ===Denoising &amp; Farfield ASR===<br /> <br /> * Lasso-based dereverberation obtained reasonable results<br /> :# Found some specious problems with frequency-dependent Lasso.<br /> :# Proposed full frequency Lasso &amp; full frequency-temporal Lasso.<br /> :# good performance was obtained with F-dependent Lasso<br /> :* Near data: 10.79 -&gt; 10.35 (lamdba=0.05)<br /> :* Far data : 40.53 -&gt; 35.65 (lambda=0.15)<br /> <br /> ===VAD===<br /> <br /> * Some discrepancy between CSLT results &amp; Puqiang results<br /> [http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=wangd&amp;step=view_request&amp;cvssid=207]<br /> <br /> :* check if the label is really problematic<br /> :* check if short-time spike noise is the major problem (can be solved by spike filtering)<br /> :* check if low-energy babble noise caused mismatch (can be solved by global energy detection)<br /> :* test noise data trained model<br /> <br /> ===Speech rate training===<br /> <br /> * Some interesting results with the simple speech rate change algorithm was obtained on the WSJ db<br /> [http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=wangd&amp;step=view_request&amp;cvssid=268]<br /> <br /> * Seems ROS model is superior to the normal one with faster speech<br /> * Need to check distribution of ROS on WSJ<br /> * Suggest to extract speech data of different ROS, construct a new test set<br /> * Suggest to use Tencent training data<br /> * Suggest to remove silence when compute ROS<br /> <br /> <br /> ===Scoring===<br /> <br /> * hold<br /> <br /> ===Confidence===<br /> <br /> * Implement a tool for data labeling<br /> * Finished extraction of two features: DNN posterior + lattice posterior<br /> <br /> <br /> ==LM development==<br /> <br /> ===Domain specific LM===<br /> <br /> h2. G determinization problem re-open. <br /> <br /> h2. NUM tag LM:<br /> <br /> 27h JS test: 20.16 vs 20.19<br /> 2h JS test: 17.48 vs 17.49<br /> <br /> h2. Analyze the property of the tag LM: (1) random NUM should obtain better performance; (2) other words are not seriously impacted.<br /> <br /> <br /> ==Word2Vector==<br /> <br /> ===W2V based doc classification===<br /> <br /> * Initial results variable Bayesian GMM obtained. Performance is not as good as the conventional GMM.<br /> * Interest group setup, reading scheduled every Thusday<br /> * Non-linear inter-language transform: English-Spanish-Czch: wv model training done, transform model on investigation<br /> <br /> <br /> ==RNN LM==<br /> <br /> * New toolkit from Thomas obtained.<br /> * Prepare WSJ database, re-test RNN.<br /> <br /> <br /> ==Speaker ID==<br /> <br /> * Second model done<br /> <br /> ==Emotion detection==<br /> <br /> * initial performance obtained<br /> [http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=wangd&amp;step=view_request&amp;cvssid=271]<br /> <br /> <br /> ==Translation==<br /> * Failed due to out of memory <br /> * Re-start the training due to some errors in grid</div>

Cslt