git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4697 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4696 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: Removing binary lattice-to-ngram-counts.cc from latbin that was accidently added along with merge from sandbox/language_id

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4695 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4694 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: Merging in i-Vector language identification updates from sandbox/language_id. Missing egs/lre07/v1 in Previous commit.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4693 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4692 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

sandbox/language_id: Merging trunk into sandbox again. Trying to fix mergeinfo problems, when merging this sandbox into trunk.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4691 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4690 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4689 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4688 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4687 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4686 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

sandbox/language_id: Removing change to egs/lre07/v3/run.sh used for testing that shouldn't have been committed.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4685 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

sandbox/language_id: Changing phonotactic example in egs/lre07/v3 to use the LRE07 closed-set criteria

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@4684 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4683 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4682 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: a lot of minor fixes, cosmetic changes and revisions (most significantly: making conjugate gradient more robust.)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4681 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: modifying gpu-based nnet2 training code to do the I/O and input-data formatting in a separate thread from the main training thread (substantial speed improvement).

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4680 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4678 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4677 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk,nnet1: adding tool 'feat-to-post', Converts features into posterior format, is used to specify NN training targets when training autoencoder.



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4676 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4675 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4673 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4672 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4670 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: add the example script local/online/run_nnet2_perturb_speed.sh in egs/swbd/s5b which tries to perturb the data in the nnet training.



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4665 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4663 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4661 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4659 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4657 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4656 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

Removing the egs/rm/s4 recipe. It's based on the older generation 's3' recipes, and now there are better examples using free data in Kaldi

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4654 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4653 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4651 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4647 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4643 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4642 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

The change is in how the PER is computed. Previously there was optional silence with no output symbol, and forgiving of silence deletions. This was however slight cheating (Traditionally in TIMIT, the silence is scored).

The change consists of:
- removing the optional silence from he lexicon (see call of utils/prepare_lang.sh in run.sh)
- removing the forgiving of 'sil' deletions from scoring (see change in local/score.sh)
- updating the RESULTS

From now on the PER scores will involve no cheating and will be comparable to experiments in the literature.



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4639 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4638 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4637 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8