Å®ÉúСÊÓƵ

A COMPUTER system which can recog nise continuous speech, launched by
researchers from Edinburgh last week, may form the basis of a personal computer
controlled by voice. The American computer company Apple, famous for its
MacIntosh range of personal computers controlled with a ‘mouse’ which moves
a pointer around the screen to select various icons, are planning to adapt
some of the approaches used in the new system for its own computers.

Mervyn Jack, director of Edinburgh University’s Centre for Speech Technology
Research, says that work with Apple began several months ago and that an
agreement on collaborative research is expected soon.

John Lewis of the Mac Users Group, an independent body that represents
people who use Apple’s personal computers, said that if the system can overcome
the hurdle of recognising continuous speech it would be a ‘very significant
development’, widening the MacIntosh’s ‘already friendly interface’.

The hardware in Edinburgh’s system, known as ‘Osprey’, consists of two
standard computer boards which plug into the back of an IBM personal computer.
The first converts the voice signal into a digital form and then analyses
the signal to pick out the key frequencies of speech. In this simplified
form the signal is passed to the second board. This holds four transputers,
self-contained computers which combine processing, memory and communications
on a single piece of silicon.

The transputer board converts the key frequencies into ‘phonemes’, the
characteristic sounds of speech of which there are 44 in the English language.
It then tries to identify the pattern of phonemes into the words of its
vocabulary and then checks the syntax of the sentences to see that it makes
sense. The developers of the system claim it can understand any voice speaking
clearly in English. It is reasonably successful with urban Scottish, they
added, and they are currently working on additional software which will
improve its understanding of regional accents.

Osprey’s developers claim it can be adapted to suit any application.
Users would type in a working vocabulary of 300 words. Osprey then looks
them up in a 20 000 word dictionary of phonemes and it is ready to use.
A major advantage of Osprey is that it requires no laborious training. Other
systems often need to be tutored by each user.

Jeremy Peckham, head of the speech unit at the British software house
Logica, believes that speech control of computers would be very slow compared
with current methods such as Apple’s ‘mouse’ and icons. But combining these
with speech control might be a different story, he added.