Home
SphinxTiny-0.7
SphinxTiny-0.5


Berkeley Institute of Design

Mobile Speech Recognition Engine

We have adapted CMU's Sphinx3.x open-source large-vocabulary continuous speech recognition system to support use on mobile computing platforms.  We are using a fixed-point MFCC front-end, and have accelerated parts of the Gaussian evaluation in assembly. This significantly increases the recognition engines's performance on devices that only support soft floating point calculations. 

We provide two versions of SphinxTiny:

• SphinxTiny-0.7 is based on Sphinx3-0.7, is designed for use on Linux and Linux Mobile platforms, and is currently being developed.
• SphinxTiny-0.5 is based on the older Sphinx3-0.5 and is designed for use on Windows Mobile devices.

Seth Horrigan
eomer At cs.berkeley.edu

John Canny
jfc At eecs.berkeley.edu