git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3526 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

The support for it had been already included into the 4-anydecode script


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3522 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk/babel/s5b) First version of the improved decoding pipeline. Does not handle the OOV search yet.
For the time being, it supports the PEM and UEM (from CMU) segmentation. Moreover, it includes the
SEG segmentation (Kaldi-BABEL native), which is currently being finalized
Another changes is the ability skipping extra features (especially related to search). This is
useful as the "quick" baseline can be run routinely and the "extra" only when needed.



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3521 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3520 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: for language-id setup adding an implementation of Shifted Delta Cepstra (SDC) features and corresponding unit tests.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3519 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3518 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3516 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3515 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3514 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3513 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

The merge improves WER by about 0.5 % abs and the ATWV by about 0.3 % abs


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3512 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

Fixes to extend_lexicon.sh: 1. added OOV rate computation as diagnostic information, if a text file is provided; 2. extended phone mapping repository to a..zA..Z0..9; 3. changed num_sent_gen as the total number of sentences to generate; 4. added encoding when processing words

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3511 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3510 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

so we have the same resolution during the whole interval (which wasnt the case when using inv-power,
where the resolution was reciprocical )


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3509 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk/babel/s5b) Adding a script for confusion matrix generation (for use, e.g. with the proxy kw search)


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3508 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

The interface is regarded stable for now.


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3507 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

this flag is "optionably deletable", i.e. the penalty will be incurred only for
a stray insertion, not  a stray deletion. This gives approximately 0.2% WER 
improvement on Zulu


git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3506 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3505 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

Changes to lower level scripts for proxy keywords: 1. decomposed generate_proxy_keywords.sh into a language dependent part and language independent part; 2. improved the ATWV performance

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3504 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

* analyze-counts refactored, accepts symbol table to print counts
* nnet-loss.cc - added sanity check to dim in input posteriors



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3503 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk/babel/s5b) Ignore the return code of the cleanup code. Othervise the 'set -e' will cause failing the script after rm returns non-zero status code

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3502 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk/babel/s5b) Reverting the previous commit as it was including other changes not related to the subject of the commit

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3501 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk/babel/s5b) Reverting the previous commit as it was including other changes not related to the subject of the commit

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3500 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3499 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk/doc) Changing documentation about getting Kaldi. SF recently seems to disallowing the svn URI access, so we suggest the https one

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3497 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: Changing openfst failover in tools/Makefile to point to my website (since it was currently failing over to a different alias of the same site)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3496 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3494 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3493 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: HTK conversion tool, adding support for 3column arcs, representing arc with implicit "unit weight" equivalent to "0,0,".



git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3492 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3491 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3490 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: BABEL scripts: adding some scripts for lexicon expansion to handle OOVs, and changes to G.fst preparation to handle probabilities for these (associated changes to run-1-main.sh to come); removing some scripts that should not have been in local/.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3489 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3488 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: wsj setup: In decode scripts removed option --max-arcs from being passed to gmm-latgen-faster, where the option was previously removed.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3487 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: speaker-id setup: Added a condition 0 to score_sre08.sh which gives us the EER over all conditions.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3486 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

(trunk/babel/s5b) Fix for a pathological case when the train_nj was specified higher than number of available audio files

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3484 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3483 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

various changes related to the determinization with phone insertion. This usually gives us another 20%~40% speedup for determinization on our Babel setup.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3482 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

mpe training parameters for Babel fullLP; limitedLP needs re-investigation as we didn't run that after fixing the normalization stuff.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3481 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3479 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8