Re: [Accessibility] Thinking Further About Speech Recognition

[Eric S. Johansson]

 On 7/26/2010 10:04 AM, Samuel Thibault wrote:
> Christian Hofstader, le Mon 26 Jul 2010 09:44:21 -0400, a écrit :
> > So, I propose that we launch a speech reco user interface and
> > pre-processor project that is 100% libre. The different sub-projects can
> > talk to libre engines and be entirely endorsed by FSF/GNU.
> I just wanted to remind the existence of the CMU Sphix engine under a
> BSD license.

And, as you probably saw another message (and I apologize for being too tired to 
say this politely) but the  Sphinx engine is a ()^&^%*&^ toy for grad students. 
We looked at it in the mid-to thousands as part of the open-source speech 
recognition initiative and spoke with a Sphinx numeral four developer who 
confirmed our assessment which was that it IVR only. Very small vocabulary to 
keep up recognition speed and accuracy. I refer you to the other message/rant.

Samuel (and others) please please please spend some time looking on the research 
for large vocabulary continuous speech recognition systems. They have a 
different architecture from systems like Sphinx. I can't articulate it at the 
level of a freshman in college which means I don't totally understand but, go 
read up on the systems please. Eventually I will do the same. :-)

I will try to get a request off  to some friends in the speech recognition world 
for good textbook recommendations.