News

Chattering PDAs will be normal this year

Fancy ball-by-ball commentary of a sporting event, whispered into your ear, by your Personal Digital Assistant? It could happen this year; you'll be able to get pretty fluent speech out of your pocket computer in a couple of months, if Scansoft's plans for "UltraCompact RealSpeak" go to schedule.

The breakthrough is to get a basic, compressed synthesis vocabulary into just 16 megabytes of memory, says European general Manager, Peter Hauser.

Then, the voice can be generated from just a few bytes of Ascii text.

Current versions of voice-based software - like the recently launched PDSay - require 64 megabytes or more. The Compaq iPaq pocket computer is almost the only one capable of being expanded to this point, and of running the software, which was launched earlier this month (February).

Hauser is currently doing a European road show to persuade all the old customers of Belgian-based Lernout & Hauspie (L&H) which Scansoft bought out of its financial collapse last year. He's hoping that new products like Dragon NaturallySpeaking version 6, will inspire software developers and resellers.

NaturallySpeaking is a dictation product. It will be some time before dictation is available for mobile devices, says Hauser; but text-to-speech products don't require anything like the same processor power, nor storage capacity.

In the old days, L&H used to sell itself as "text to speech, speech to text, text to text, and speech to speech" - hoping to have a full translation service for live voice available. Text to text - translating English to French from a computer document, for example - is now commonplace, with many Web sites offering the service.

But the dream was of having someone speak English into a phone in London, and have German emerge from another phone in Hamburg, or Japanese from one in Tokyo. L&H even did some demos purporting to be "prototype" simultaneous translation packages, but eventually, was forced to concede that the technology was still beyond it.

Even so, one component of the dream was to have an entire vocabulary - every word - of a language, embedded in a computer. And since modern phones are computers, the dream stays alive. It would be very useful if a phone could be made to speak the contents of an SMS, for example.

With modern pocket computers, the dream approaches sensible reality. The phone and the pocket computer can be linked - and increasingly, this can be done over wireless, with Bluetooth.

With the phone connected to the Internet via GPRS, for example, and the PDA able to "wake up" when the phone asks it to, it becomes theoretically possible to get a constant stream of audio over the air, without spending a fortune, and without losing the ability to use the phone for its main purpose - phone calls.

For example, cricket or baseball fans could listen to a ball by ball commentary. It would be delivered in a pretty robotic voice - RealSpeak isn't going to fool anybody into thinking it's a human being - but it will be perfectly possible to understand it. And unlike a radio show, this would be receivable around the world with minimal bandwidth, as a simple text stream.

The new L&H isn't as wildly ambitious as its predecessor. There are no promises of software which will watch TV for you and transcribe news programs in real time. Nor is there any suggestion that simultaneous translations over a phone can be done by a standard PC. What there is, rather, is a hope that the parallel worlds of voice recognition (the Dragon brand) and of optical character recognition (OCR) in which Scansoft dominates with its Omnipage brand, can share research and technology.

In that context, the idea that a PDA could "talk" seems far more realistic.