Dragon Voice Recognition Computer Software
Learn About The Newest Digital Technology
A while ago, Tom Kroeger wrote and asked about applying the new MMX technology to speech recognition, and how long it would be till improvements showed up to make use of this technology. I'm no longer in the speech industry, but my guess would be quite a while (one or more years for real improvements). The mmx instructions would be somewhat helpful for speech, but to really take advantage of them will require rewriting code and changing data structures. It's going to be a while before compilers support MMX (as far as I can tell), and no one likes to write in assembler. Of course, it could be done. But because the MMX can't perform arbitrary multiple operations, only operations on data all packed into the same 64 bit space, data structures will probably have to change, too, which means relatively large code changes, not just inner loops. Of course, someone in marketing is going to say "We need to use the MMX to get sales" and someone in engineering will put in a small hack that makes a small improvement, and tell the sales guy, who will put it in big letters all over the box. So, when the box comes out, ask how much faster the thing actually runs. Meanwhile, I wouldn't hold my breath, and I wouldn't run out to buy one... --Joshua Goodman ----- goodman@eecs.harvard.edu Joshua Goodman -- Computer Science Graduate Student studying NLP On Fri, 17 Jan 1997, Tom M. Kroeger wrote: > > With all the talk about performance I thought this question > might be of interest. > > I was just reading and article, in the January " Communications of the > ACM" that talks about the details and performance improvement of the new > mmx Pentiums. The quick and simple explanation of it is that > mmx technology provide several new instructions that can manipulate > (e.g. add multiply) several 8, 16 or 32 bits numbers at the same time. > For example I could process four samples of data from my 16 bit sound > card all in 1 instruction vice 4. This method of doing parallel > computation commonly known as single instruction multiple data (SIMD). > > This article noted significant performance increases for multimedia > applications such as graphics and speech (e.g. voice recognition), > the one catch is that in order to see such performance improvements > a program must specifically make use of these special instructions. > > My question is if anyone (especially those people on the list > working for voice recognition developers) knows if any of the > voice recognition products (e.g. dragon, Kurzweil and voice type) > have plans to make use of this technology? If so will > such improvements come as a patch or and upgrade, and > his there any official statement of time when such improvements > could be expected? (Ok, maybe I'm asking for a lot here, > but I'll take what I can get). Thanks, > > > tmk > > ----------------------------------------------------------------------- > Tom M. Kroeger Pray for wind > Graduate Student, UC Santa Cruz \ Pray for waves > e-mail: tmk@cs.ucsc.edu |\ and Pray it's your day off! > http://www.cse.ucsc.edu/~tmk |~\ > (408) 459-4458 |__\ > (408) 427-2001 home ,----+-- > >