donw@zehntel (Don White) (02/15/90)
<This would not mail directly. It might be of general interest anyway, so ...> In article <1348@crash.cts.com> you write: >In-Reply-To: message from Dalamar@cup.portal.com > >I don't have all the info yet, but I'm learning how to use it right now...my >boss wants me to have a *DYNOMITE* demonstration worked out in the next week >doesn't work, you have to unplug it if you wish to talk with out VoRec going >berserk and doing stuff like opening CLIs, etc. It also helps if you are in a >very quiet room...while the mic isn't that sensitive, it will pickup outside Hi, I used to play with voice recognizers alot! Back in my pre-Amiga days I was the voice recognition 'expert' for my company. My favorite box was made by a company called Votan. Their box was a statistical pattern recognizer. i.e. You would 'train' the patterns you wanted to recognize into the box's memory and when you spoke it would perform a statistical waveform comparison to find a match with a learned pattern. I found that some voices have better characteristics for consistent recognition. This was due in part to the varying properties of microphones. Each mike filtered the voices just enough so that I found it best to work with each operator and find the best mike for each of them. The catch to the statistical method is that errors become more likely as your number of recognition patterns increases. So, the best rule of thumb is to limit the number of patterns as much as is possible. In addition, I did a Pascal program interface for this project which allowed a user to back out when the rare errors occured, by issuing a verbal cancel. My Program had the ability to verbally activate different sets of patterns. This meant that any individual set was fairly small but you could set up a fairly complicated verbal menu system with a high degree of accuracy. If pulling the microphone out is the only way to turn it off, you probably have a broken mike switch. My purpose in sending you this note is twofold. One, if my experiences are of use to you, you are welcome to them. Second, I am curious about the capabilities of VoRecOne. 1. Does it work in the same fashion as what I was used to? (ie. does it use learned words or phrases to compare against? or is it using phonetic recognition?) 2. Does it allow you to control voice recognition through program control? Also, what's your project? I am curious. Thanks, Don White Box 271177 Concord, CA. 94527-1177 zehntel!donw