I've used Sphinx 4 with multiple gram files and hacked together an interface for Google STT. Sphinx 4 worked well, but the paradigm is you need to know the target words before recognition happens. Google is very impressive, but it is "in the cloud" - so your device would have to be online.
Here's a link of my findings....
I was hoping someone had actually used it and could act as a Sensei here. (In other words, I did google it before posting.) I've worked with some other CMU speech software (I got their pronouncing dictionary when I was working with the Babblebot and played with adapting the allophone sets to the ginsing library.) In any case, does this mean I get to be the LMR pioneer? The documentation talks about turtle programming so I don't think it's completely novel.
My intended implementation is an openCV-able self-mobile web browser, essentially: tell it "come here" and ask it a question, have it google the answer and read it back to you. Some of that seems to be built into Chrome on my Wintel box-anyone using a Linux chrome dist. that has the little microphone in the search box? That would save a lot of time if it was available. Oh well, I'll know soon enough. The Beagle Board is on its way.
... As soon as I saw this post I started looking ... found all of them sphinx, and the others.
Tryed poketsphinx, got it compiled and all, added the dictionaries ... it seems to be working but my accented english is not doing it any good :) So I did some digging this thing goes deep man I have trouble understanding what to do with it to train it. YMMV I guess.
I run Linux but the opensource chrome does not have the mic.
Thanks TinHead! You're the best. What did you compile it on? Are you using a Raspberry Pi?
I compiled it on Archlinux which I use currently but I guess it would work on Ubuntu too it's pretty much posix stuff.
Basically you need this http://sourceforge.net/projects/cmusphinx/files/sphinxbase/0.7 installed first then this http://sourceforge.net/projects/cmusphinx/files/pocketsphinx/0.7 which makes use of it.
I would suggest you to use the latest sources as older binary releases seem to depend on old style /dev/dsp which is long depreciated.
Once you get them compiled and installed you will also need the "dictionaries" or models for your language http://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/
That's as far as I got ...
... to hearing more about this. Blog?
sphynx, but, it is not plug and play. It is pretty in depth as far as I understand it.