News: Downloadable PDA Demos
Introduction
![]()
Technical Description
![]()
How To
![]()
Documents
![]()
Product Brief
![]()
White Paper
![]()
Supported Platforms
![]()
Consulting
![]()
License Terms
![]()
NFL Demo
![]()
Downloads
![]()
Secure Travel Demo
![]()
FAQ's for Users![]() &
Developers
![]()
Registration Form and License Agreement
![]()
SDK Access
![]()
Client Access
![]()
Feedback
![]()
Other Sprex Products/Services
![]()
Exit
|
 |
 |
 |
ANSRTM
achieves state of
the art accuracy with reasonable engine memory footprint and CPU
requirements on medium-sized grammar tasks. For example:
- a grammar in the form of a 4000 word lattice is recognized using a
1GHz CPU in 8M of virtual memory at 15x faster than real-time.
- A Sprex customer, Vicor, has achieved as high as 100% accuracy
(1364/1364 words correct) in laboratory tests with speaker adaptation
on a small numbers-and-workflow grammar.
- ANSR inherits the
accuracies achieved by speech engineers from Cambridge University,
authors of HTK and HAPI, which were sold by Entropic, Inc., until
Entropic was purchased in 1999 by Microsoft.
- CU-HTK systems regularly
achieve the lowest error rates by a margin of 2 standard deviation
relative to the average error rates of submissions of the leading
speech laboratories in the world on the US government NIST benchmark
tests for speech recognition accuracy, 1 standard deviation lower than
the second-lowest submission. See the NIST website for CU-HTK results
on Hub4 and Hub5 tests, for example.
- Independent tests by Sehda Inc.,
have shown the ANSR engine's accuracy to be superior to that of IBM's
recognition engine.
|