I was wondering if there is any difference in pocketsphinx performance when
using fsg or n-gram language models.
n-gram language model can be also converted to fsg, for example with SRILM.
The difference in performance depends on how well FSG or LM matches the target
language so it depends on a tools you are using to construct the grammar or
the language model, not on pocketsphinx. Most of the tools used to construct
grammar work only for specific inputs. For example lm toolkit is effective to
construct grammar from free form text with large vocabulary. The equivalent
fsg could be hard to construct though possible.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I was wondering if there is any difference in pocketsphinx performance when
using fsg or n-gram language models.
What's different between them? (I know the syntax is one! ;) )
Thanks,
Amin
n-gram language model can be also converted to fsg, for example with SRILM.
The difference in performance depends on how well FSG or LM matches the target
language so it depends on a tools you are using to construct the grammar or
the language model, not on pocketsphinx. Most of the tools used to construct
grammar work only for specific inputs. For example lm toolkit is effective to
construct grammar from free form text with large vocabulary. The equivalent
fsg could be hard to construct though possible.
I am using cmulmtk for language model, and I really don't know anything about
fsg, so I was worried that I am missing something.
Thanks for the reply