Thanks dhdfu!
Actually I was exploring a way to convert a CDHMM(1stream) to SCHMM(4stream) in order to skip AM training steps, but I found out that this is not possible logically.(Am I right?)
So I decided to re-run the lovely SphinxTrain :)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The released version of PocketSphinx doesn't support CDHMM models, so you will have to use a nightly snapshot or get the code from Subversion. Make sure you use the snapshot/SVN version of SphinxBase as well.
You won't need to convert anything, it ought to just work (though it will possibly be slower than Sphinx 3.6). Make sure that you pass -feat 1s_c_d_dd, though.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have a CDHMM model trained with 1s_c_d_dd and working well in s3.6,
Is it possible to use it in pocketsphinx anyway?
When I use it in pocketsphinx, the following error occurs:
FATAL_ERROR:
"\pocketsphinx-0.4.1\src\libpocketsphinx\s2_semi_mgau.c", line 1150: ./am/means: #codebooks (650) != 1
Should I convert anything?
Thanks.
Thanks dhdfu!
Actually I was exploring a way to convert a CDHMM(1stream) to SCHMM(4stream) in order to skip AM training steps, but I found out that this is not possible logically.(Am I right?)
So I decided to re-run the lovely SphinxTrain :)
Hi,
The released version of PocketSphinx doesn't support CDHMM models, so you will have to use a nightly snapshot or get the code from Subversion. Make sure you use the snapshot/SVN version of SphinxBase as well.
You won't need to convert anything, it ought to just work (though it will possibly be slower than Sphinx 3.6). Make sure that you pass -feat 1s_c_d_dd, though.