Features

Tegic one year from launching voice-pen-button texting - first impressions

The idea that you could enter text into your phone by a method other than triple-click is tempting, but also, sounds far-fetched; so when Tegic invited 3GSM delegates to see a demo, scepticism was not uncommon.

The demo was reasonably convincing, all the same - but it isn't slick. This is a very pre-production demo, it takes a powerful PocketPC to run it, and it is completely un-sellable to a phone maker at this stage. But here's what it does:

1) it lets you switch from texting with the keypad, to scribbling with a handwriting recogniser, to picking letters off a picture of a qwerty keyboard with only a vague approximation of accuracy (Sloppy Type)- and yes, finally, you can enter by speaking the words.

2) you can go back to something you created with words, and edit the text with a pen, or a keypad, or sloppytype.

3) believe it or not, the voice recogniser works, already, and will almost certainly work fine in a year or so when this starts appearing in handsets.

In short, this is very proof-of-concept. Said Eric Collins,  VP Sales and marketing: "We have to integrate the cores. If any OEM tried to bring this to market today, you'd have standard T9 and its core, you'd have Sloppy Type, its core and engine, and you'd have handwriting plus its core and engine, and finally, speech - and integrating them all together would mean they probably wouldn't build a phone before we had an integrated solution read out of the box."

The speech recognition engine hasn't been finalised, yet, even; "we hope to announce which partner we're going with in four or six weeks," Collins estimated.

The good news is: it should be out on any PocketPC quite quickly - within 12 months, estimates the software department. "That's the first platform because we have the tools to do it on that. Then we'd take the other platforms."

Which other platforms? They weren't keen to commit themselves, but allowed one to gain the impression that PalmOS wouldn't be next on the list; probably Symbian.

Plans are to take this engine and integrate it with the applications platform. Tie in MapQuest, for example, and you can talk your way through the zooming and scrolling needed to find a place to eat or meet. That will make this more than just a text-editing tool. It will be part of the UI.

Impressions: too soon to say how well speech will integrate into the package. It's set up for an American accent; that doesn't mean it doesn't work with anybody else, but it also doesn't mean you'd be impressed if you had a London dialect; and the demo was very "pre-canned" in the sense that it clearly wasn't being given all the options.

That also means that it isn't clear what the user interface will be. The product manager, Lisa Nathan, used a red button (the normal button for hanging up a call) to invoke the speech recogniser - and obviously, you wouldn't want the recogniser to start trying to guess what you're talking about whenever it felt like it, and filling up the machine's memory with garbage. But would you want to accidentally hang up on an incoming call?

It's a non-trivial question, still to be resolved: for the typical text user, accustomed to pressing the "3" key three times to get the letter "F" it is confusing enough to switch to Tegic predictive text and finding that you don't know what the "magic key" is to switch to capital letters, for example, or how to find a comma instead of a full stop.

More modern phones give the user clues - a list of possible words pops up half way through the word. Tegic says the next product will advance this concept considerably, and be more friendly; as you speak, you'll see a choice of words that sound similar to what you're saying.

This can work, but experience with continuous dictation software suggests it's not easy for learners. It's very hard to concentrate on what you're trying to say, and at the same time be reading what you said two words ago. And the recogniser on the handheld is very unlikely to be able to do dictation; individual words, probably.

We'll see...