The easiest way to use Speech Recognition with Asterisk isprovided by Unimrcp project

http://www.unimrcp.org

For example you can use CMUSphinx through unimrcp pocketsphinx plugin.

http://code.google.com/p/unimrcp/wiki/PocketSphinxPlugin

MRCP is a standard protocol for the speech recognition and it provides both great flexibility of the deployed solution as well as interoperability with the commercial products.

--- (Edited on 8/19/2012 18:38 [GMT+0400] by nsh) ---

In addition to nsh's valuable suggestion of MRCP you can also use AGI (Asterisk Gateway Interface) directly. The idea is that you use agi in a script to pass audio to and from the Asterisk server. Once the input audio is recorded on the server in the form of a wav or other file, you just pass the file to Julius and get the response back, enunciated using TTS over the phone again using AGI functionality.

I've used this technique successfully with a number of languages, grammars, audio models, devices etc. For audio examples see http://www.it4gh.com/wp.

One of the downsides is that an audio model created with a desktop mike on a local machine in a user-dependent fashion does not work well over AGI unless you boost your audio training with samples from the telephone context which of course takes time and effort. To get round this quickly you need a larger user-independent model, which is where CMUSphinx can help.

--- (Edited on 8/20/2012 2:19 am [GMT-0500] by colbec) ---

--- (Edited on 8/21/2012 1:01 am [GMT-0500] by arielberg) ---