The "Files" section contains local copies of software libraries on which
Kaldi depends, as well as tools and data files needed for the work
of some of the example scripts.
Note that these files are not meant to be downloaded manually by users.
Instead the respective installation and example scripts will download them
- openfst-1.2.10.tar.gz - the version of OpenFst library, currently
used in Kaldi
- sctk-2.4.0-20091110-0958.tar.bz2 - a local copy for NIST Speech
Recognition Scoring Toolkit
- wsj_ascii.tar.gz and wsj0-train-spkrinfo.txt - data used by the WSJ recipe
- rm1_cepstra.tar.gz - a local copy of the cepstra for a subset of RM
corpus, originally distributed by CMU
- RM_G.fst - pre-built grammar(G) level transducer used in egs/RM/s4 recipe.
- waves_yesno.tar.gz - data needed by the example scripts in egs/yesno/s3.
- sequitur-model4 - pre-trained model for Sequitur letter-to-phone
tool. Used by egs/voxforge/s5.
- online-data.tar.bz2 - data needed for the online decoding demo in
For more information about the Kaldi toolkit please visit http://kaldi.sf.net