Date: Thu, 19 May 1994 18:58:38 GMT From: dp@world.std.com (Jeff DelPapa) Subject: Re: articulate systems Organization: The World Public Access UNIX, Brookline, MA Newsgroups: sci.med.occupational In article <2rfsuk$7jk@mudraker.mtholyoke.edu>, Evelyn Wright wrote: > >I'm in the process of choosing a voice recognition text-entry program, >and I'm wondering what people think of the ones they've used, particularly >Articulate's, since I have not read ANY reviews of theirs. The full-bore version of articulate is still in the "about to release" state. (I have seen it demonstrated) -- basically it is dragon, one release behind (both in price and algorithm) It has the advantage that it doesn't require DSP, just a normal soundcard (and thus will run on the new mac powerbooks "blackbird", finally a battery option) The first version is not a general keyboard replacement, it accumulates text in a separate window, which you then paste into the application. They promise to fix this. Lots of the people here use DragonDictate. (this text was generated that way) It was the first system available, and once they had competition, they have been pretty aggressive about upgrades, both in function and price. It is a MS-DOS system, with all the warts so engendered. There exists a workable kludge for unix/X users, and the result is a pretty stable system. They are promising a windows version shortly (and I have reason to think they mean it this time). The sales reps get mixed reviews, the Boston one has a clue, but I am told the LA one doesn't. The tech support organization is helpful, the sales organization is less so (I have gotten answers to my tech questions, but my attempts to buy a replacement foam bit for my microphone resulted in unreturned telephone calls). Another alternative is Kurzweil. I have tried one briefly. It isn't compatible with the unix screen export kludge, but seems to work quite well. (it was better than v1 dragon, but v2 dragon matched it) It required less training time than dragon, but I hated the "beep in your ear" part of the interface. There support organization gets mixed reviews (I found them confused when I called them), and Kurzweil companies have not been noted for their stability. It is a DOS product, they have announced a windows version. IBM has been re-marketing a version of DragonDictate under their voice-type label (a 7,000 system). They also have their own system under OS/2. It is labeled as a 32,000 word system, but in reality it has a 20,000 word active vocabulary, 2,000 words for the user, and optional (at extra cost) specialized vocabulary modules. (radiology, legal, etc). Like the articulate system, it accumulates text into a special window which you have to use pointer gestures to paste into the application. Unlike articulate (which promises to fix this), this is the intended design behavior of the system, which they exploit by using a delayed recognition system. If they are in doubt as to which word is meant by an utterance, they will delay recognition until you have said one or two more words, as context analysis is much easier if you have words after, rather than trying to do it based only on the previous words. If you turn this "feature" off, the recognition accuracy is unspectacular. In common to all systems. They are discrete utterance (pauses between words), require partial training and identification of the speaker (all support multiple sets of voice models), and allow free text (not just commands). In all cases, the faster the machine the better. Usual disclaimer: I don't work for any of the companies involved. I paid full price for the system I use. I can't even get on the beta tester list. The only system I have used for more than 4 hours is the Dragon.