Besides, what I am really trying to do is experiment with speeding up the code, so I would like to disable usage of the language model entirely, not just set its weight to a low (or zero) value. Is there some other flag or trick I can use?
Much thanks,
Mike
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
(I apologize if someone has already answered this question in the Forums, but I have been searching for a while for an answer and cannot find one.)
Is there some way to disable use of the language model when using the Sphinx 3 decoder? I have tried setting -lw 0, but I get an error:
INFO: subvq.c(109): Precomputing Mahalanobis distance invariants
FATAL_ERROR: "lm.c", line 585: lw = 0.000000e+00
Besides, what I am really trying to do is experiment with speeding up the code, so I would like to disable usage of the language model entirely, not just set its weight to a low (or zero) value. Is there some other flag or trick I can use?
Much thanks,
Mike
You still need language model to be able to restrict search, you can't search for any word in the universe.
But you can use either simple unigram model with only a list of words (see an4 demo for example) or you can use fsg (-op_mode 2 -fsg test.fsg).