Update from original #4
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…shared script in utils/data, and delete the unnecessary scripts
… (improves results)
Bug fix in steps/paste_feats.sh
…g with nnet3; refactoring the nnet3 decodable code.
Adding nnet-latgen-faster-parallel program for multi-threaded decodin…
- using GetVerboseLevel(), - avoiding 'WriteIntegerVector' for writing to KALDI_LOG by introducing: 'operator<< (std::ostream, std::vector<T>)' in kaldi-error.h
…cleanup; and nnet3+no-cleanup results
…ed (#960) the results. Fixed a bug in steps/nnet3/align.sh when supplying online-ivector-period option to nnet3-align-compiled
… and add the output to RESULTS files for tedlium/s5_r2 and ami/s5b
nnet1: redesigning LSTM, BLSTM code,
Scripts for model info
updating the 'rm' (B)LSTM results,
* Remove gawk dependency from s5c swbd1_prepare_dict.sh (for #1143) * Suppress warning about non-existent lexiconp.txt * Check for 2,435 or 2,438 data files in SWB corpus
Add the capability to enforce a maximum change per component in nnet3, at the minibatch level. This CHANGES THE DEFAULT BEHAVIOR OF EXISTING SCRIPTS. nnet3 and nnet3+chain config-generation scripts will now by default apply a max-change at the per component level. You can override this (for back compatibility) by adding --max-change-per-component=0.0 --max-change-per-component-final=0.0 to the make_configs.py scripts. However, we don't recommend doing this unless it is part of an existing experimental setup that you need to keep consistent, because we found that the max-change-per-component is slightly helpful (0.1% or 0.2% absolute in our experiments).
* Fix gawk-specific features problem in prepare dict scripts
* every third field as separator * using internal get_ctm.edits.py * remove other files have copies in interval * fixed error messages * removed duplicated make_one_biased_lm
a) Fix get_ctm_edits.py to be not too greedy with id's. It would go wrong if an id was a prefix of another id b) Add lattice-1best command in ctm generation for the non-position-dependent-phones case
#1146) * Some changes to the unk-model implementation to ensure determinizability and to keep L.fst small. * Fix a pathname in error message in make_unk_lm.sh (thanks: Jasper Ooster), comment out results in local/run_unk_model.sh
#1157) * fix the scripts by putting lattice-1best * Copy the descriptions from Dan and add a example for lattice-align-words-lexicon #1155 (comment)
…isher data. Resolves issue #1139.
The latest version of sw-ms98-dict.text from http://www.openslr.org/resources/5/switchboard_word_alignments.tar.gz contains the following header on line 1: file: $SWB/data/dictionary/sw-ms98-dict.text This ends up in lexicon0.txt, which causes utils/validate_dict_dir.pl to fail with the following error: --> ERROR: phone "$swb/data/dictionary/sw-ms98-dict.text" is not in {, non}silence.txt (line 10399) This in turn causes utils/prepare_lang.sh to fail. Update dict.patch so that it removes the file header from sw-ms98-dict.text.
sendMail -> SendMail SendMail() removed from train_rnn.py (already declared in nnet3_train_lib)
…run for the first time (#1160)
For wider perl version compatibility
* env.sh is under tools not tools/extras. * Fixed a couple typos.
Print correct location of env.sh
* Make sure orig2utt will be sorted, otherwise decoding will later fail * Don't use a syntax that causes 'Experimental keys on scalar is now forbidden' error with Perl 5.23 and later
Depending on how LDC2002T43.tgz is unpacked, the transcripts might be under 2000_hub5_eng_eval_tr. As requested in #1169, we append it to tdir if that directory exists.
fix cpp style guide dead link
* added recipe for swedish * requested changes made * requested changes made * exit 1 added and copyright * removed dict_prep.sh * changed sph2pipe
* Reverberation based augmentation recipe added for swbd. Improves over the best TDNN recipe. BLSTM/LSTM recipes pending. Added minor modifications to reverberate_data_dir.py to enable inclusion of the original data directory.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.