Chips Promise to Boost Speech Recognition
From: CNet - 08/23/2006
By: Stephen Shankland

Carnegie Mellon University researcher Rob Rutenbar believes the key to making
speech recognition a practical reality is using a custom computer chip. At
this week's Hot Chips conference in Palo Alto, CA, Rutenbar said, "It's time
to liberate speech recognition from the unreasonable limitations of
software." A custom processor could be used for speech recognition similar to
the way in which special-purpose hardware has been used for graphics. Speech
recognition software continues to be plagued by speed limitations and power
demands. The "in silico vox" project at Carnegie Mellon consists of custom
ASICs and FPGAs. Rutenbar provided a videotaped demonstration of Carnegie
Mellon's speech recognition technology using a low-end FPGA, and it
recognized short sentences about two times as fast as researchers were able
to speak them and matched the university's Sphinx speech recognition software
in accuracy. First-generation custom chips are expected to be about twice as
fast as the rate of regular speech for a 5,000-word vocabulary. One custom
chip is being developed to work at 10 times the spoken rate, and there are
plans to reach speed factors of 100 and 1,000. 

Read the entire article at:
http://news.com.com/Chips+promise+to+boost+speech+recognition/2100-1008_3-6108417.html

Links:
Rob Rutenbar
http://www.ece.cmu.edu/directory/details/150/

Carnegie Mellon Engineering Researchers to Create Speech Recognition in Silicon
http://www.cmu.edu/PR/releases04/040913_speech.html

Sphinx speech recognition software
http://www.androidtech.com/html/speech-recognition-free-softwa.php
http://sourceforge.net/projects/cmusphinx

Hot Chips Conference 2006
http://www.hotchips.org/hc18/main_page.htm