kaldi

История

Dan Povey 4377ba39d1 Changes to READMEs git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@1356 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8		2012-09-16 17:34:52 +00:00
..
s1	Small bug in where wer results are written to. The wer results needs to be written to a location that the code in line 62 can read from.	2012-01-12 20:30:10 +00:00
s2	Merging various small script changes from sandbox	2012-09-09 19:29:17 +00:00
s3	Committing s5 scripts for sgmm2; not fully tested but mostly finished.	2012-06-24 17:49:17 +00:00
s4	egs/rm/s4/ cepstra download fallback to Kaldi's "Files" at SourceForge,	2012-07-03 07:25:47 +00:00
s5	Merging various small script changes from sandbox	2012-09-09 19:29:17 +00:00
README.txt	Changes to READMEs	2012-09-16 17:34:52 +00:00

README.txt

About the Resource Management corpus:
    Clean speech in a medium-vocabulary task consisting
    of commands to a (presumably imaginary) computer system.  About 3 hours
    of training data. 
    Available from the LDC as catalog number LDC93S3A (it may be possible to
    get the same data using combinations of other catalog numbers, but this
    is the one we used).

Each subdirectory of this directory contains the
scripts for a sequence of experiments.  Note: s3 is the "default" set of
scripts at the moment.

  s1: This setup is experiments with GMM-based systems with various 
      Maximum Likelihood 
      techniques including global and speaker-specific transforms.
      See a parallel setup in ../wsj/s1
      This setup is now slightly deprecated: probably you should look
      at the s3 recipes.
      
  s2: This setup is experiments with pure hybrid system.

  s3: This is "new-style" recipes (now superseded by s5)

  s4: A recipe based on freely available subset of RM data, distributed by CMU

  s5: This is the "new-new-style" recipe.  It is now finished.
      All further work will be on top of this style of recipe.  Note: 
      unlike previous recipes, this now uses the same underlying
      scripts as the WSJ recipe.