Friday 5 July 2002

Pic of the day: Does this look like to dragon to you? (The box.)

Dragon Naturally(?) Speaking

We interrupt our studies in theology to proclaim to you the good message: The Dragon has arrived! OK, that sounds kind of bad in context. (Though it may actually make sense to those of you who have been reading Robert Jordan's Wheel of Time books, but that's not what I mean.) The arrival is actually Dragon Naturally Speaking speech recognition software, which I ordered from Amazon UK some days ago.

Except for the cheesy name, it is actually quite good. I first installed it on my portable HP running Windows XP, using the microphone that came with the program. This was not the great success. I suspect the microphone is to blame. The program runs better on my desktop machine under Windows ME and with the microphone I bought a couple years ago for voice chats. The fan of my desktop machine makes a constant buzzing sound, but this does not seem to affect the performance at all.

Between the two machines, I have spent the whole evening training the program. In the end I was afraid it would not recognize my normal voice, only the rusty one. I am not normally a great talker: I spend most of a conversation listening (unless someone brings up one of my favorite topics at least) and besides I don't go out much. So talking for hours on end tends to ruin my voice. I have the same problem when I have to teach at work, but this is not very often anymore.

***

So what do I think? Well, this entry is written with my microphone and the occasional helping finger. The program operates with blocks of text, consisting of all words you say without a pause. This can be many or a few,depending on your confidence and lung capacity. The software tries to interpret one such block at a time, and writes it to the Windows program of your choice. If it guesses wrong (and it usually does at first), you have the option to press the minus key on the numeric keyboard. This will bring up a list of alternatives. Quite often what I tried to say is listed as one of the alternatives, even if the first choice looks like something very different. By choosing the correct alternative (which I can do by speaking), I not only update the text but also trained the program to guess better in the future.

Given that English is not my mother tongue, and given that I read a lot more than I listen, I am actually impressed. This version is for U.S. English, which I barely even understand when spoken. So being able to actually use it productively after one evening is a pleasant surprise.

Speech recognition is not yet suited for classrooms of homes with children. But given that this is not even the latest version of the program, nor the most advanced, the day might actually arrive when we can speak to our computers in a relaxed and natural way. (Provided, of course, that we aren't sucked into a black hole before then. Stop the mad scientists at CERN!)

(Oops! I just discovered that it is difficult to dictate while laughing – because the program wrote "mad scientists at CRFH". Proof that it has indeed read those months of archives that I fed it to expand its vocabulary!)

And with that, goodnight, and thanks for your attention!