LanguageModelHowTo

[Anonymous]

Hello,
I was experimenting with Pocketsphinx and mostly following the tutorial and
other How To pages. On the LanguageModelHowTo page after creating your own
language model and dictionary using the lmtool it suggests trying the
following command:

pocketsphinx_continuous -hmm /usr/local/share/pocketsphinx/model/hmm/en_US/hub4wsj_sc_8k \
    -lm 8521.lm -dict 8521.dic

When I used this (substituting my own .lm and .dic files) the output
frequently was blank or (null) for the hyp variable. However, when I removed
the -hmm argument and used:

pocketsphinx_continuous  -lm 8521.lm -dict 8521.dic

(again, with my own .lm and .dic files) pocketsphinx recognized the words in
my dictionary without any error. In the pocketsphinx_continuous man-page it
says this argument is required and when looking at the rest of the output I
saw that no default value was listed.

I am unsure about what this argument does. Also, how does it relate to
sphinxTrain and training the rm1 and an4 acoustic models? I understand that
the training allows your model to more accurately recognize speech for your
particular application, but I am unsure how the program incorporates your
trained model.

Thanks for your help

[Nickolay V. Shmyrev]

pocketsphinx recognized the words in my dictionary without any error.

I don't believe into that. The
/usr/local/share/pocketsphinx/model/hmm/en_US/hub4wsj_sc_8k is a default model
which is loaded in both cases. You probably need to test it a little bit more.

am unsure about what this argument does.

It loads acoustic model from specified location

Also, how does it relate to sphinxTrain and training the rm1 and an4
acoustic models?

It's unrelated

I understand that the training allows your model to more accurately
recognize speech for your particular application,

No, it's misconception

but I am unsure how the program incorporates your trained model.

Acoustic model is provided with above option and loaded on startup