GitLab

trunk/nnetbin : Introducing length constrain of 1 min on segments that are used for mini-batch frame cross-entropy DNN training. This will avoid possible out-of-memory issues by discarding the very long segments which are infrequent.


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2943 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

fix a bug - the last argument was checked by ClassifyRSpecifier, but it should be the first argument, since the last is a W specifier


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2939 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2938 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2932 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: code change that should fix memory leak in bin/latgen-faster-mapped-parallel [fix other branch of the code, omitted first time]

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2927 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2926 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk/nnet : alignment script can optionally generate output in lattice format (handy for getting word alignments)



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2914 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2907 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2906 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2903 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2896 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2894 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2892 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2888 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

This causes warnings (which are supressed), but shouldn't affect the functionality of the program. The code in question is part of non-essential UI portion of tthe program and may be removed entirely, if more problems arrise.


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2875 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk:egs/hkust update on chinese character error rate (CER) scoring script. Add --useword2charmap option and comments.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2872 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2869 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: Making compile-train-graphs{,-fsts} report num-succeeded more accurately, and output no empty FSTs

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2868 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

Fix to nnet example script run_5b.sh, and to train_sgmm.sh RE --ciphone-list (bug RE context-dependency).

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2864 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2847 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2838 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: Merging sandbox/dan2.  Code and script changes RE my neural-network training setup, including refactored scripts.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2830 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/dan2@2829 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

sandbox/dan2: Script progress RE neural net training (mostly minor) and SGMM RE quinphone (doesn't help)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/dan2@2828 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

was done to follow the convention used in the other scripts


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2827 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/dan2@2826 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2825 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2823 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/dan2@2819 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2810 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2809 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2808 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2807 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2806 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2805 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2804 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2794 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2791 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2789 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk/src/nnet : bugfix in on-the-fly frame-shuffling, very long segments are now truncated correctly (hard limit on leftover to 1/2 of cachesize). Thanks to Nagendra for finding it!


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@2788 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8