Is it possible to use an interpolation of two language models with Sphinx 3.5? I would like to use an interpolation of a word trigram and a class trigram.
OK, if I interpolate using the CMU LM toolkit, I get a LM that contains both classes and words. But, all my words in the word trigram are also class members in the class trigram.
So the question now is: can sphinx handle the case where a word is in the LM 'as is', and also as a class member? I.e. word probability is a sum of its explicitly defined probability and the probability computed via its class?
Thanks in advance.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Is it possible to use an interpolation of two language models with Sphinx 3.5? I would like to use an interpolation of a word trigram and a class trigram.
Also, is the Sphinx 3 language model control file (given with the -lmctlfn parameter to decode) documented anywhere? The best documentation I found is this example: http://fife.speech.cs.cmu.edu/sphinx/models/lm/turtle-class-lm/
OK, if I interpolate using the CMU LM toolkit, I get a LM that contains both classes and words. But, all my words in the word trigram are also class members in the class trigram.
So the question now is: can sphinx handle the case where a word is in the LM 'as is', and also as a class member? I.e. word probability is a sum of its explicitly defined probability and the probability computed via its class?
Thanks in advance.
Try CMU LM toolkit interpolate. -Arthur
You probably need to write your own program. Hopefully, later version will support that functions. -A