News

Are Pocket PCs fast enough for speech recognition?

by Guy Kewney | posted on 05 February 2002


Speech recognition - trying to make it work on pocket PCs is a challenge which has been taken up by ScanSoft's Dragon division. But do these little machines have the raw processing power to do a good job?

Guy Kewney

You talk to it, and it obeys. At least, that's the idea. For around thirty pounds, you can download Dragon PDSay, a software package for your PocketPC which will make your control of the machine hands-free, says ScanSoft.

Still untested, the software comes with plenty of promise: "With Dragon PDsay, Pocket PC users can control operating system selections and application interfaces using natural voice commands and can have e-mail, calendar and contact information read aloud automatically," says the release.

PDsay works with the standard microphone and speakers built into Pocket PC products.

"ScanSoft's speech and language technologies bring new functionality to small-footprint and limited-display devices," said Mark Erwich, Director of Marketing, Europe at ScanSoft.

If it works as claimed, PDsay will pleasantly surprise many. The claim is that it "demonstrates how state-of-the-art speech and language capabilities" can be delivered to small-footprint devices and embedded applications doesn't sound quite the same as "new functionality" - and observers say the proof of the pudding will want to know just how "state of the art" it can be.

When speech recognition first became possible on desktop computers, it was recognised that you needed a processor at least as powerful as the early Pentium-based PCs - in the days when they could reach 120 MHz. Technology has advanced, certainly - but are today's pocket PDAs really that powerful?

Scansoft has produced a trial user: Andy Ai, a Web and IT Consultant with InnerLink Strategic Solutions Inc. He says he relies on ScanSoft's Dragon PDsay technology to quickly access critical customer information and important e-mails. "Whether I'm at my desk or on the road, I rely on PDsay to quickly and accurately search for contact information without having to type in a name, or view the screen. I just say the name of the person I need to contact and PDsay reads the phone number back to me."

PDsay claims to do speaker-independent speech recognition and has "a flexible command vocabulary that helps customers instantly use and benefit from the application without training."

Experts say that this almost certainly means that the device works with a very limited vocabulary. "You couldn't get speaker-independent recognition without training, unless you were restricting your recognition set to a very small number of words," said one developer. "The pocket device doesn't have the storage capacity or the processing power to handle normal speech in real time."

The software understands and acts upon simple natural commands such as "Next Appointment," "Read e-mail" or "Start Internet Explorer."

Key features include:

* Information retrieval by voice (contacts, e-mail, appointments, tasks)

* Speech notifications

* Automatic text-to-speech reading of emails

* Voice recognition and control of standard built-in Pocket PC functions and applications

* Instant command and control through speaker-independent speech recognition

* Zero-training through natural command vocabulary

ScanSoft's Dragon PDsay is immediately available in English as a downloadable purchase from www.scansoft.com/PDsay.