Blog

Speech recognition on Tablets, and running shoes than run off by themselves

Speech recognition. It seemed like a good idea! - what I needed, yesterday, was a day without typing. I recognise the symptoms of RSI (repetitive strain injury) and they were unmistakably starting in my left wrist. So I finally got around to invoking the Tablet PC.

This particular Tablet is rather nice; it has an array microphone, which can be set for narrow, medium, or wide range. The idea is that this way, you can get it to listen only to you, or to two people, or a group around the table.

What you really should not try to do, is set it up to do speech recognition. But, baffled by enthusiastic hype, that's what I did.

Don't do this.

First off, Tablet Speech recognition isn't that good. Oh, definitely, it's better than the first copy of Dragon Naturally Speaking which I tried out in the early days of the first Intel Pentium processors - but that's to be expected. But it requires consistency. That means "get a headset." With an array microphone sure, you get far better quality of sound input, but it varies as you move your head. A headset with a built-in microphone, however, makes sure the mic is always the same distance from your lips, no matter how you move about.

And the headset mic doesn't pick up the noise of the PC speaker.

The trouble is, every time the speech recogniser completes a sentence, it beeps. The beeping is picked up by the microphone, which seems to be determined to translate it as speech, and the result was chaotic. And when I tried to dictate a paragraph complaining about this, I found that the word "beeping" isn't one Recogniser recognises - and I tried, many times:

"The speech recognition software itself is making the penalizes. The King the pain the King the pain the V. V. - there's no way of really get this thing to say 'beating' it just doesn't believe BPA's word ."

Having spent a good hour "training" the recogniser to work with the array microphone, I abandoned it, and switched to the headset. That involved another hour of re-training. At the end of that, things were improved, but I still couldn't recommend it as a way of avoiding the keyboard.

I'm trying to get hold of the latest Dragon version, now in the hands of Nuance. It's probably best not to discuss that. I showed up for the Dragon V9.0 launch on the day I was given (I thought) and found nobody there, and when I enquired was told I was a day late. "But we'll send the stuff," they said.

What I wanted, of course, was a review copy, to write about. What they meant was: "We'll send you a completely useless colour print out of three volumes of PowerPoint." D'oh!

The actual software does exist. I read an excitable review of it in the New York Times where David Pogue claimed it really did recognise his pronunciation of "infinitesimal zithers."

I'll probably spend the rest of the month wondering what he really said. What I won't be doing, is testing it, because apparently, they only made enough copies of the software "for the people who showed up".

Excuses, excuses. Only two weeks ago, I got exactly the opposite excuse from Apple, who launched a wireless running shoe with Nike. I showed up for that, and saw row upon row of evaluation wireless units, iPod Nanos and running shoes. "Those are for people who said they were coming," said Apple.

I used to work in PR. I know exactly what happened, in both cases: the stuff earmarked for invited journalists but not collected on the day got snaffled by company staff.

Tags: speech recognition, Tablet PC, Nike, RSI