Path: utzoo!utgpu!watmath!clyde!att!mhuxu!m10ux!rgr
From: rgr@m10ux.UUCP (Duke Robillard)
Newsgroups: comp.sys.ibm.pc
Subject: Re: Voice recognition/speech synthesis products
Summary: Covox: good synth, mediocre recog.
Keywords: voice speech
Message-ID: <807@m10ux.UUCP>
Date: 30 Dec 88 15:37:23 GMT
References: <300@holin.ATT.COM>
Reply-To: rgr@m10ux.UUCP (Duke Robillard)
Organization: AT&T Bell Labs, Murray Hill
Lines: 60

In article <300@holin.ATT.COM> doc@holin.ATT.COM (David Mundhenk) writes:
>I have been very interested in speech synthesis and voice recognition
>.... November PC Tech Journal has an ad in the back from
>"COVOX,Inc." (TM?) listing "Speech Products" including "Synthesizer-
>$79.95", "Digitizer-$89.95" and "Voice Recognition-$49.95".

I bought all the Covox stuff (mostly cause it sounded cool and
was cheap.  And I love I/O stuff.  anyone got a track-ball they
wanna unload?  Or any footswitchs?  Has anyone bought that funky
flight simulator plane control thing that clamps on your desk?)

When I got it, it was in two parts: a speech synth and a
digitizer/voice recog thing.  Looks like they've unbundled the
voice recog. software.

The voice synthesizer is pretty good.  It just a 8 bit DAC in 
a module that plugs into your parallel port (without hurting
your printer) and a little speaker.  There's a device driver
for putting out digitized voice files (you can do this from
inside a program, so all your error message can be vocal)
and a third-party program that'll read text. Each bunch of
digitized info that goes out has to be less than 64K, which 
is only a couple of secs (8 secs, I think, at the worst
resolution/sampling rate) but you can send one after another
without much gap between 'em.  The text reader has that
computer voice, but is pretty good.  It'll get words wrong
occasionally, but not too bad.

The digitizer is a 1/2 size  card with an 8 bit ADC and a 
microphone plug.  It's good for generating files for the
synthesizer to speak.  From what I can see, the voice 
recognition software is kinda lame.  It's easy to use,
but not particularly good.  It's definitely speaker-dependent
(i.e. recognizing one individual doesn't help in recognizing
anyone else) and you really have to speak the same way you did 
when you were training it.  This is also a device driver. To
use it you send a "train" command and some training data for the
each word you want to recognize (64 max).  Then you pass it
other data and it returns the number of the word that matched,
or some error code.

TI has just come out with a new voice recog. chip (I read
about it in EE Times yesterday) so maybe you should wait on
that part.  There are gonna be speaker independent voice
activated toys soon, so can a good PC board be far away?

There's a cool digitized voice file editor that comes with
both the synth and the digitizer.  This is useful for 
doctoring recordings (:->)  I can see it now:  
"Did you kill him?  speak into my computerized recorder."  
"I did not!"
(edit, edit) REPLAY: 
"Did you kill him?  speak into my computerized recorder."  
"I did!"


-- 
|  Duke Robillard                                                        |
|                                  |                                     |
|  ARPA:     rgr@m10ux.att.com     | UUCP:     {backbone!}att!m10ux!rgr  |
|  or maybe: m10ux!rgr@att.att.com | BITNET:   rgr%m10ux.uucp@psuvax1 (?)|