The other day I started talking to my computer. I told it a number of fables by Aesop, read it something by Bill Gates, and even dictated Bertrand Russell's The Problems of Philosophy to it. When I was done, my computer understood me -- at least enough to write these words down as I spoke them.
Voice recognition software is one of the Holy Grails of current computer research, with the potential, when it is perfected, of revolutionizing the way that we interact with machines. But, like many cutting-edge innovations, voice recognition has notoriously been so unreliable that it has been all but useless. Recently, however, both IBM and Microsoft have been quietly making strides in the field, proving that voice recognition may now be a working option.
For several years now, the standard software has been the Naturally Speaking system, but the newest version is plagued with problems that make it not only incompatible with some word processing programs but also difficult to manage on the computer itself. IBM's Via Voice was, until Microsoft's Tablet PC, the most accurate software available, and it works with any Windows program, although it's pricey at almost $200. Microsoft's own software comes integrated with Office XP, and only works with programs like Word and Excel. The most recent - and greatly improved - version of Microsoft's speech recognition software is integrated with Tablet PCs, however, and can work with almost any Windows application.
With any of the programs, some training is necessary. For example, my cat Pangur's name was initially understood as either "anger," or "painter." After teaching a Tablet PC his name's spelling and pronunciation, however, it comes out correct every time. At other times, the program understands without clarification. If I say "to, too, and two," Via Voice understandably writes "two, two, and two;" but it chooses the correct words when I dictate the sentence: "This is too cool; send two of these to me." And it's amazing how liberating it is to tell the computer to open up your folder of favorite Web sites and watch as it visits the one you name. Of course nothing is perfect, and plenty of errors still crop up. But at this rate, it won't be long before software developers will have conquered the problem of voice recognition. Then they can move on to perfecting how humans communicate with each other.
Publication date: 05/08/03