Making PocketSphinx Less Resource Efficient

[Tikiboy]

Hey guys, i was just wondering if its possible to play with pocketsphinx
configurations and parameters to make it less efficient resource wise, without
hurting the accuracy too much.

I need to make resource usage measurements of pocketsphinx under different
configurations. I am looking for ways to cause pocketsphinx's speech
recognizing processing to consume more memory and CPU is possible.

http://cmusphinx.sourceforge.net/wiki/pocketsphinxhandhelds
I was referencing the above link and found that by playing with a particular
param i can increase memory usage, however i am unsure about the remaining
parameters.

Would it be possible for someone to point out which param is the best to
modify to achieve:
- a higher memory consumption
- a higher cpu utilization

A breif explanation on why these params have the intended effect would also be
great. Based on what i know, a likely way to increase memory usage is by
expanding the search tree used by pocketsphinx, however i am unsure which
param is responsible for that action.

Thanks alot!

[Nickolay V. Shmyrev]

Hello

In the default configuration most of the search space is reduced by the beams

-beam       1e-48       Beam width applied to every frame in Viterbi search (smaller values mean wider beam)
-fwdflatbeam    1e-64       Beam width applied to every frame in second-pass flat search
-fwdflatwbeam   7e-29       Beam width applied to word exits in second-pass flat search
-lpbeam     1e-40       Beam width applied to last phone in words
-lponlybeam 7e-29       Beam width applied to last phone in single-phone words
-outlatbeam 1e-5        Minimum posterior probability for output lattice nodes
-pbeam      1e-48       Beam width applied to phone transitions
-pl_beam    1e-10       Beam width applied to phone loop search for lookahead
-pl_pbeam   1e-5        Beam width applied to phone loop transitions for lookahead
-wbeam      7e-29       Beam width applied to word exits

You can make all the beams smaller to disable pruning. I.e. wbeam could be
1e-100 instead of 7e-29. Then search will be slower and more accurate.

To learn what beam is check any text book on speech recognition, for example
this lecture:

http://www.isip.piconepress.com/publications/courses/msstate/ece_8463/lecture
s/current/lecture_34/

[Tikiboy]

Thanks, ill take a look at the lecture to learn more.