| |
  |
| Cramming Speech Recognition Software Into PDAs |
|
 |
|
posted by Editor on Thursday October 11, @09:17AM
|
|
|
|
|
This article in today's New York Times (free registration required) reveals some efforts going on to bring speech recognition software to handheld computers. This is, of course, particularly challenging since handhelds trade off processing capability for size and power consumption, and accurate speech recognition is notoriously compute-intensive. The article discusses IBM's Personal Speech Assistant (PDF) and Voice Signal's ELVIS, which can be used to dictate email messages to Compaq's iPaq.
|
|
|
|
< Recreating The Real World In Virtual Space | Open Source Stroke Translation Library > | |
|
|
|
Don't have an account yet? Go Create One. A user account will let you customize the site's content according to your preferences. It will also allow you to moderate the comments of other users.
|
|
|
|
|
|
This discussion has been archived.
No new comments can be posted.
|
|
|
|
|
by Anonymous Coward on Thursday October 11, @09:26PM EST (#1)
|
|
|
|
|
When I wrote a paper on speech-driven interfaces in 1998, I dictated the first paragraphs using Dragon Naturally Speaking on a pentium 166 with 64MB of RAM. The iPaq runs a 206Mhz StrongArm processor, and some models have 64MB of memory (though not all available for runtime applications).
It doesn't seem an incredible stretch to get OK recognition on a handheld. Of course, we've seen how successful uptake of speech-driven software has been [not very] - 95% accuracy still means making corrections every 20 words. Speaking at 160 words a minute that's 8 mistakes per minute to fix.
In the near term, I think dedicated applications (like speech driven menuing) that only have to recognize words within a relatively small set are more useful, since they can get 99.9% accuracy.
|
|
|
|
|
|
|
