DEAF people and language students could learn to produce difficult speech
sounds more accurately using a technique being developed at the University of
New South Wales in Sydney. People learning a new language find certain vowel
sounds hard to master, and vowels can be particularly tricky for people with
hearing problems.
The Australian system displays a moving two-dimensional image on a computer
screen of the mouth shape and tongue position of a speaker pronouncing a sound
properly. At the same time it monitors the position of the student鈥檚 tongue and
mouth as they try to pronounce the sound, and displays it next to the target
image. Students try to make their image match that of the target. So even
without hearing it, they will eventually pronounce the sound very accurately.
鈥淚t鈥檚 a bit like a video game,鈥� says Joe Wolfe of the development team.
The system works out the position of the speaker鈥檚 tongue and the shape of
their mouth by testing how they affect a specially synthesised sound signal
generated just outside the speaker鈥檚 mouth. 鈥淭hat sound is affected by the
position of the speaker鈥檚 mouth and throat, and from that we can work out how
open the mouth is and where the tongue is,鈥� says Wolfe.
Advertisement
The idea behind the technique is not new. In the late 1980s, various teams in
the US looked into it, but did not have the necessary computing power to produce
an effective system. Now Wolfe believes the new device has great potential.
Alison Purcell, a lecturer in speech pathology at the University of Sydney,
agrees. 鈥淚t sounds like a good idea,鈥� she says.
Other feedback techniques for teaching pronunciation all have problems when
it comes to vowels, says Wolfe. 鈥淭here are technologies to help completely deaf
people to speak. However, they have all had limitations in their performance
with vowels and that鈥檚 where we鈥檝e made what we think is a significant advance,鈥�
he says.
The team hopes that its technique could prove particularly useful for Asian
students learning European languages. 鈥淨uite a lot of languages, for example
Japanese, have only five vowel sounds,鈥� says Wolfe. 鈥淓nglish has 12 and French
16, so this is where our technique could be very important.鈥� The researchers
have tried out the prototype device with English speakers learning to pronounce
French sounds, and with people temporarily deafened by white noise that masks
the sound of their voices. They will reveal detailed results at the Australian
Acoustic Society Conference, to be held in Brisbane later this month.
Wolfe says there is still a lot of work to do. 鈥淚t鈥檚 a lab tool at the
moment. The technological performance is impressive, and it appears to be
successful, but how it will go in a classroom, we can鈥檛 say.鈥�
