[VoIP] More fun with speech recognition

[Russ Price]

I've been playing around with another speech recognition system, based 
on the CMU Sphinx2 engine.  A handy little package with an AGI 
application and a Perl server program connects Asterisk to Sphinx.

Advantages:
* Does not rely on an outside server that could disappear or hide behind 
a paywall at any time
* Software is free

Disadvantages:
* Very difficult to set up
* Fantastically complicated recognition engine with documentation best 
suited for computer speech gurus (not I!)
* Voice recognition from a carbon mic (like my trusty WECO 2500) is 
spotty at best - condenser mics are much better
* Likely to have problems with UK English - or even with southern US accents

I'm running the speech software on one of my Ubuntu MythTV systems, 
since it has a very fast CPU (Core 2 Duo E8400), and Ubuntu has 
precompiled sphinx2 packages.

The SPREC package that makes Sphinx2 usable with Asterisk is at 
<http://uc.org/read/sprec>.  The CMU Sphinx programs themselves are at 
<http://cmusphinx.sourceforge.net/html/cmusphinx.php>.

I'm definitely *not* going to do any handholding regarding how to set 
this all up - it's highly experimental, and very subject to tweaking - 
especially if I can somehow figure out a way to make it work better with 
carbon transmitters. It is *not* a drop-in replacement for app_asr in 
any way, shape, or form.

Oddly, my North Electric 6H6 performs slightly better than the 2500 with 
this setup - but my Uniden cordless phone works best.  I found that a 
yes/no confirmation step was absolutely vital due to the recognition 
problems, at any rate, and I run separate recognizers for digits and for 
yes/no questions.

I've set up a CNET robo-operator at +1 442 9901 if you want to test it 
out.  It expects the full CNET number including the country code.

I've noticed that the LumenVox engine is now available for Asterisk, 
with a "starter kit" for $50 - and prices that go up steeply for greater 
capabilities.  <http://www.lumenvox.com/partners/digium/asterisk.aspx>

	Russ
	+1 442 7877