Voice key was recently released, it contains
 
1) the Sphinx-based voice dictation project, "Voice Keyboard" 
2) a set of helper scripts for making language models 
3) sphinx3_livesegment, which is pocketsphinx_continuous (tty-continuous.c), ported from pocketsphinx to sphinx3, except that it also listens on stdin for a small set of commands which can be used to dynamically switch the language model or change some configuration parameters 
4) a couple of language models trained on corpora capturing my own computer use (one of them contains some word frequencies and one of them contains keystroke frequencies) 
5) a set of scripts and instructions for creating and using acoustic training data 
 
Voice Keyboard has a vocabulary of about 17,000 words roughly corresponding to the vocabulary of lm_giga_20k with proper nouns removed. It supports typing individual keystroke using codewords such, for example HENRY ECHO LIMA LIMA OSCAR. It also has codewords corresponding to most movement and modifer keys, for example CONTROL LEFT ARROW. Accuracy on the keystroke codewords is decent. Accuracy on the full vocabulary is not great but not unbearable either -- I have been using Voice Keyboard to do most of my typing for the past two months. 
 
http://sourceforge.net/projects/voicekey/ 
 
See the project documentation (in the download; particularly HACKING and also the doc/ directory) for more detailed information on items 2-5.

--- (Edited on 7/28/2008 6:06 pm [GMT-0500] by nsh) ---