Commit Graph
152 Commits
Author SHA1 Message Date
David Huggins-Daines eb14c19ef0 port to python3 and test 2020-11-09 16:36:16 -05:00
David Huggins-Daines 6fcbf2b612 Port to python3 and add a test for s2mfc 2020-11-09 11:40:14 -05:00
David Huggins-Daines c78b4d0220 Port to Python3, add context managers 2020-11-09 10:14:03 -05:00
David Huggins-Daines 9f39f3cb8d Convert to Python3, fix Unicode issues, and test 2020-11-09 09:38:06 -05:00
David Huggins-Daines 84cf1295b1 switch to setuptools and change the package name to be the model name, also to reflect the fact that these are not really used in SphinxTrain 2020-11-09 09:37:40 -05:00
nshmyrev 96c92bc076 Fixes mistake
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/sphinxtrain@13114 94700074-3cef-4d97-a70e-9c8c206c02f5
2015-10-24 12:15:25 +00:00
nshmyrev 0e5d9b14a2 Numerically stable likelihood computation
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/sphinxtrain@12918 94700074-3cef-4d97-a70e-9c8c206c02f5
2015-04-12 18:14:22 +00:00
mbait 69f83361fc Fixx invocations of ngram_prob.
ngram_prob does not support variable arguments anymore.
It should be a list of strings.


git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/sphinxtrain@12380 94700074-3cef-4d97-a70e-9c8c206c02f5
2014-03-20 09:34:38 +00:00
nshmyrev 59cf62db04 Don't use E_WARN_SYSTEM, it has been removed
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@12038 94700074-3cef-4d97-a70e-9c8c206c02f5
2013-11-07 18:59:59 +00:00
nshmyrev 2be16edcdb Changes quantize_mixw to use sendump class, not it reads sendumps properly
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@11368 94700074-3cef-4d97-a70e-9c8c206c02f5
2012-05-20 10:11:18 +00:00
nshmyrev b77932f224 Added scripts to prune and quantize mixture weights for pocketsphinx
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@11367 94700074-3cef-4d97-a70e-9c8c206c02f5
2012-05-20 09:45:11 +00:00
nshmyrev fc489b4ed0 CMLLR estimation requires full covariance. Still not sure
if we need unshifted covariance or shifted one.



git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@11349 94700074-3cef-4d97-a70e-9c8c206c02f5
2012-04-13 18:13:03 +00:00
nshmyrev 4f91f3074f Removed some more unused headers and fixed distcheck too
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@11271 94700074-3cef-4d97-a70e-9c8c206c02f5
2011-11-23 07:24:35 +00:00
nshmyrev c220c68091 Install python scripts and use them for training
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@11261 94700074-3cef-4d97-a70e-9c8c206c02f5
2011-11-16 18:49:18 +00:00
nshmyrev 397a8145fd Suprisingly, distcheck works
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@11238 94700074-3cef-4d97-a70e-9c8c206c02f5
2011-09-30 22:33:58 +00:00
nshmyrev 3d79cb43df Merged automake port
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@11236 94700074-3cef-4d97-a70e-9c8c206c02f5
2011-09-30 21:34:57 +00:00
nshmyrev 6663bceafb Proper normalizer for gaussian. Fixes issue
https://sourceforge.net/tracker/?func=detail&atid=301904&aid=3299965&group_id=1904

Thanks to Vassil Panayotov



git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10963 94700074-3cef-4d97-a70e-9c8c206c02f5
2011-05-12 12:53:48 +00:00
nshmyrev 0adc313da1 CMLLR estimation by Stephan Vanni <svanni@vecsys.fr>
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10783 94700074-3cef-4d97-a70e-9c8c206c02f5
2011-02-17 14:34:37 +00:00
nshmyrev 2abc104308 Adds some methods required for rescoring which were removed accidentally I suppose
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10642 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-11-14 16:24:24 +00:00
dhdfu bb791f2553 add support for creating a single pronunciation dictionary in force alignment
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10587 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-11-01 22:08:17 +00:00
dhdfu a29cda9232 add a pruning option to lattice error calculation (but you should use fstutils)
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10586 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-11-01 16:48:04 +00:00
nshmyrev 91523eda18 Sendump update to be closer to converting pocketsphinx sendumps. Not
really usable now since it doesn't read 4-bit mixtures properly



git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10579 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-10-22 12:55:12 +00:00
lqin 254072d227 Implementation of MMIE training
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10163 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-06-24 20:03:30 +00:00
dhdfu 37c6338d43 add an accumdirs() function to s3mixw so norm can be written in python
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10107 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-17 15:43:08 +00:00
dhdfu b88c181888 switch to adjacency list (actually hash table) representation for fsgs, binary compatibility be damned. now sphinx_jsgf2fsg will actually compile good sized grammars (it still produces lousy output but openfst can fix that easily)
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10082 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-04 07:17:22 +00:00
nshmyrev dea6417998 Raises error when there are problems with building MLLT matrix
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10063 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-03 13:38:17 +00:00
dhdfu f337ea6968 Update and pull things from my bzr repository. Lattice rescoring,
with and without FSTs, which may be useful, plus a little script to
generate composed class language model FSTs.


git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10060 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-03 04:48:55 +00:00
dhdfu 87faadd1dc add support for building and evaluating FSTs from class-based language models
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10059 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-03 03:33:26 +00:00
dhdfu cc093e13f0 Turn on keyword substitution for Revision:
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10058 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-03 01:05:39 +00:00
dhdfu 47c6947a69 Tweak the old lattice_error implementation, and add a new one based on
openfst, which also handles compound words properly.  Also add my
Python FST utilities, for lack of a better place to put them.


git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10057 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-03 01:04:00 +00:00
dhdfu 1cc7ad14c9 fix typo
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10037 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-01 03:02:33 +00:00
dhdfu 7710d8d6fc only strip class tags when calculating lattice wer
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10036 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-01 02:57:30 +00:00
dhdfu 89257747ca bring arpalm.py in line with sphinxbase module
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10035 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-01 02:57:18 +00:00
dhdfu 4334006fa4 handle zero probabilities in HTK lattices, also strip class tags in baseword()
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10032 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-01 00:50:27 +00:00
dhdfu 209166fe5f handle zero probabilities in HTK lattices, also strip class tags in baseword()
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10031 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-05-01 00:50:18 +00:00
dhdfu f42f28cb7b start looking for sphinxbase - it is time to put wave2feat out of its misery
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@10021 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-04-29 15:09:39 +00:00
dhdfu 5b7efbcd15 oh ... that is why self[w] = phones did not work. okay and also now we have dict unions, which was the whole reason for this adventure in the first place
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9799 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-03-07 05:02:22 +00:00
dhdfu 2db07abe32 vastly elaborate on the s3dict module
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9798 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-03-07 02:53:52 +00:00
dhdfu c6d9ed4bae test for dict and data
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9796 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-03-07 00:36:49 +00:00
dhdfu ac263db1f7 move tests to more reasonable names
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9795 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-03-07 00:33:57 +00:00
dhdfu 4100fc29ce add module for reading/writing ts2cbfn files, probably going to use it for PTM (thus furthering the creeping Pythonification of SphinxTrain)
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9638 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-01-12 03:49:05 +00:00
dhdfu fbc2a45182 Incomplete support for building up tied-mixture models from
semi-continuous models (not sure that's the right way to do it,
testing on larger data sets now...)


git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9618 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-01-06 21:55:32 +00:00
dhdfu c4208076e6 well, that sucks. python now has a standard module called "sphinx". grumble. change everything to "cmusphinx"
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9615 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-01-06 16:23:01 +00:00
dhdfu 2e84e9382c well, that sucks. python now has a standard module called "sphinx". grumble. change everything to "cmusphinx"
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9613 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-01-06 15:41:04 +00:00
dhdfu cfcf2989e9 Useful method I forgot to commit before
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9609 94700074-3cef-4d97-a70e-9c8c206c02f5
2010-01-06 03:55:26 +00:00
dhdfu 169d095581 Dictionaries are assumed to be in utf-8
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9391 94700074-3cef-4d97-a70e-9c8c206c02f5
2009-10-20 03:00:35 +00:00
dhdfu 6c39465c6d Make detecting gzip files actually work (I hope?)
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9390 94700074-3cef-4d97-a70e-9c8c206c02f5
2009-10-20 02:56:39 +00:00
dhdfu 2102a4d6bd Add a phone_id_nearest() like the one in PocketSphinx
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9296 94700074-3cef-4d97-a70e-9c8c206c02f5
2009-09-11 04:43:08 +00:00
dhdfu 3ff482982f Simple module for reading/writing pronunciation dictionaries
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9295 94700074-3cef-4d97-a70e-9c8c206c02f5
2009-09-11 04:42:43 +00:00
dhdfu b61e0411f7 Oops!
git-svn-id: svn+ssh://svn.code.sf.net/p/cmusphinx/code/trunk/SphinxTrain@9017 94700074-3cef-4d97-a70e-9c8c206c02f5
2009-02-23 18:54:44 +00:00