On (03 Mar 95) [EMAIL PROTECTED] wrote to All... GF> Date: Fri, 03 Mar 1995 10:11:40 -0500 GF> From: Gene Fender <[EMAIL PROTECTED]>
GF> Voice recognition is NOT a silly notion for the SAM. GF> I was not expecting to use it for dictation, afterall. GF> There were even some little robot toys several years GF> ago that could distinguish enough to accept about 6 GF> commands. You might also be impressed by some of the GF> PC voice rec software. If you use a limited vocabulary, GF> then the software has only to make the best match and GF> could even intelligently guess from the phrase list. I think Gene's disapeared off the list now but he's quite right! In Every Day/Practical Electronics mag' Interface articles in Aug,Sept & Oct94 they wrote about speech recognition and it's problems and workable solutions. The result was some external analogue hardware interfacing to the IBM-PC Analogue Joystick port and a compiled QBASIC program that performs the base functions of learning and recognising words in 8bit sample resolution AND a slow sample rate! The method used was to split the incomming voice into highpass and lowpass signals and rectifying and smoothing the two signals to provide envelope shape data not the actual speech istelf. These only need to be sampled at say 25 samples per-second and instead of storing the entire sample of each time the word is spoken, an average difference pattern is formed from the high and low pass envelopes and that is stored as the reference. If compiled QBASIC can do it on a PC then an Interupt driven routine *could* do it on the sam! And a dedicated M/Code program would do it easily and probably spend most of it's time waiting to read the next samples! That's why I originally mentioned an 8 channel 8bit ADC as it'd have lines to spare and would be easy enough to hook up:-) Johnathan. ... Information Super-HYPE-Way. The Marketeer's Paradise! -- |Fidonet: Johnathan Taylor 2:2501/307.9 |Internet: [EMAIL PROTECTED] | | Standard disclaimer: The views of this user are strictly his own.
