Apropos ash mentioning the wintox voice recog s/w, I've hooked the standard product to a simple wheel-bot made from a Microbric Viper kit.

In the vid you'll see a serial line going off to the left. That hooks up to a USB-to-serial converter and is attached to my laptop. The laptop is running a simple "hyperterminal"-like program and the speech-to-text is simply diverted into that.

The ASCII text  goes down the serial line and the Viper checks each text "word" it sees for the ones it knows. If the word is in its command set, it performs the operation; otherwise it plays a tune on a little speaker. Even tho I don't say anything wrong in this demo, you'll hear the tune play a couple of times — usually off-camera while I'm saying "stop" repeatedly!

This is — as you can guess — a first cut. It worked better than I expected for 30 mins of playing around  (basically, fixing the rob together).

Its main problem is the s/w on the Viper. While the motor control there is fine, the loop that checks for RS232 input misses quite a few chars coming down from the laptop at 9600 baud. That's mainly due to crappy programming on my part & not allocating enough time to reading the buffer and too much controlling the motors.

I'm going to have to learn whether the "Atom Basic" (MBasic) can handle anything like threads. It certainly can handle interrupts, but the manual seems to warn you off in case you lock the processor up.

Enjoy!

http://www.youtube.com/watch?v=KvIdbfzY3Mc

Yes, so far it's working fine. But I'm just discriminating between words at the moment, and the Microsoft black box seems to be tuned for that.

I've previously gotten speach generation (i.e. full text to speech with about 12 accents) onto a 16 mb arm single-board computer; I'm still trying to find out how to do that with the speech recog.

I had previously tried "reversing" a speech gen to get a discriminator out of it(*), but with "limited success".

(*) I.e. if the robot sounds like this when it's saying the word "abort", look for something with statistically-similar features in the audio stream from the human and just pulse that switch to "0" if you do.