Pocketsphinx and Wave File Recognition

[Anonymous]

Hello everyone,

I have been learning how to use sphinxtrain and pocketsphinx. So far I have
followed the tutorial and build an acoustic model from CMUSphinx wiki. I got
to the point where I can do the continuous recognition from the microphone
using pocketsphinx_continuous.

Now I've been wondering, what can I use pocketsphinx to recognize a wave file
instead of the microphone? Do I use pocketsphinx_continuous or do I need to
use pocketsphinx_batch. I couldn't figure out how to do it. I look into the
documentation on the folder of pocketsphinx, it seems to explain the API
instead of the actual parameter for the command-lined executable.

Any guidance, would be much appreciate. Thank you in advance!

P.S. Is there a search function for this forum? I try the search bar, but it
seems to cover the entire sourceforge instead of the CMUsphinx forum.

[creative64]

You need to use pocketsphinx_batch for this purpose. If you run
pocketsphinx_batch without any arguments, it
would give you the list of command line arguments. An example commd line
option is:

pocketsphinx_batch -hmm hub4wsj_sc_8k -dict abc.dic -jsgf abc.jsgf -adcin yes
-cepdir .\abcd\audio -cepext .raw -ctl abc.ctl -hyp exp.hyp

Where -hmm points to the hmm directory
-dict points to the dictionary file
-jsfg points to the grammar file in this case
-adcin should be "yes" as shown
-cepdir points to the directory where audio file clips are stored
-cepext shows raw format in this case, can be changed to wav
-ctl points to a control file which have the names of audio files in order they will be consumed for recognition
-hyp can be used to create a file with output log

[Nickolay V. Shmyrev]

Yup, and search is here:

http://sourceforge.net/search/?group_id=1904&type_of_search=forums

You can also use entry below the red line.

[Anonymous]

I'll try that. I appreciate your help. Thank you so much!

[Nickolay V. Shmyrev]

Btw, this functionality was recently added to cmusphinx trunk. You can
checkout from svn/download snapshot and then try

pocketsphinx_continuous -infile test.wav

[Anonymous]

It seems my version doesn't work with that command. It goes straight to the
recognition from the mic, I guess I need to get the snapshot version of
pocketsphinx then. In that case, do I need to uninstall the current
pocksetsphinx or would it just overwrite the current one? Do I need the
snapshot version of sphinxbase as well?

Also, I have an off topic question. I notice that there is a link to Ubuntu
binary on the download page of CMU Sphinx, which links to this address: https
://launchpad.net/~dhuggins/+archive/cmusphinx/+packages

How exactly is this thing work? I'm kinda new at Ubuntu and would like to
learn more about it.

Again, I appreciate your time helping out. I learn alot from this place :)

[Anonymous]

Turn out I need to get the snapshot version of sphinxbase in order to install
snapshot version of pocketsphinx sucessfully.

Anyhow, I tried

pocketsphinx_continuous -infile test.wav

But it just went straight to the usual microphone recognition again (where it
ends with Ready...). Do I need to specify anything else on the command line?

Thank you

[Nickolay V. Shmyrev]

But it just went straight to the usual microphone recognition again (where
it ends with Ready...). Do I need to specify anything else on the command
line?

Please double check that you don't have some old version and you are running
the snapshot indeed.