- no lattice re-generation after 1st epoch of sMBR
- the results are now better
- adding auto-selection of queue options for BUT

- supports arbitrary number of languages,
- uses 'multi-task' training with per-language weights (one softmax per language),
- (the decoding script not yet merged)

Fixing one more recipe. The script add-start-end.sh is in bin/ as well, so using it from there, not from scripts/

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5244 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk,nnet1,mmi : bugfix in inital data filtering in nnet1 mmi script, thanks to Angel Castro for finding it!



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5240 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: small fix to get_lda_block.sh (thanks: mozno@users.sf.net) plus a few other unrelated small changes

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5238 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5233 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5232 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5228 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5224 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5223 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

- 'blocksoftmax' is better name as it has same name as the used output component,
- we use multitask objective function, so we can see performance on individual tasks,
- also adding some older LSTM results on tedlium, not working on this recipe now,



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5222 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5218 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5214 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5205 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5204 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk) The decode scripts now fail when the scoring fails. Added --skip-scoring switch to all of them to allow skipping the scoring alltogether, i.e. to avoid the decoding failure

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5202 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk,nnet1 : bugfix in cuda-gpu detection (mmi/mpe script itself is not running on a machinge with GPU, it submits job to a GPU node)


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5201 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5195 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5193 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5191 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5189 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

- it works, but it needs further analysis, so far it did not outperform the LSTM.



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5183 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5181 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5180 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5178 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5177 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8