REPORTS last week that NASA engineers have invented a mind-reading device were exaggerated. But they are developing a system that, by monitoring the electrical signals sent to the speech muscles, can work out what people are saying 鈥� even if they do not speak out loud.
At the moment the system can recognise only a handful of words. But if it can ever get close to recognising normal language, it could provide an alternative to throat microphones for people working in places where speech can be difficult, including workers in noisy factories, pilots, divers and astronauts. It might even allow people to carry on conversations in silence or restore speech to people with a damaged vocal system.
The technology is being developed by Chuck Jorgensen鈥檚 team at NASA鈥檚 Ames Research Center at Moffett Field, California. It relies on two sensors, one on the neck and one under the chin, to measure the signals sent to the tongue and the larynx, or voice box (see 鈥淩eading speech鈥�). The sensors themselves are not selective, and pick up electrical signals from all the nerves and muscles nearby. The trick is to filter out the unwanted noise, says Jorgensen, though he will not say how this is achieved.
Advertisement
The team has trained a computer program to recognise the muscle activity corresponding to someone speaking the numbers one to ten and six other words, including stop, go, left and right. The system gets the word correct 92 per cent of the time, according to initial tests. But, like early voice-recognition systems, it can achieve this only after being trained for a particular person, and it can identify only individual words, not continuous speech.
The potential advantage of such a system over throat mikes (which cut out noise by picking up vibrations directly, rather than via the air) is that it can recognise words even when they are spoken silently. This ability to listen in on someone speaking to themselves without any facial or lip movements has been described in some reports as mind-reading. But this is not what is happening, says Jorgensen. There has to be muscle movement in the larynx and tongue, even if very slight. 鈥淚t鈥檚 like silently reading to yourself,鈥� he says. 鈥淲e are not picking up someone鈥檚 thoughts directly from their brains.鈥�
Silent reading does not just happen whenever we think in words, says Walter Thumfart, head of the Ear, Nose and Throat Clinic at Innsbruck Medical University, Austria. It is something we are very much in control of. This is easily demonstrated by reading fast, he says. 鈥淵ou don鈥檛 have a chance of enervating the muscles quick enough.鈥�
This means the Ames system could not be used to interrogate suspects. But it does raise intriguing possibilities. If the technology makes its way into everyday equipment such as mobile phones, people who learn how to subvocalise at will could have silent conversations during a film or a business meeting. The Ames team is already working on using a new type of sensor, originally developed in the UK at the University of Sussex, which requires no contact with the skin and so could be built into clothing (New 杏吧原创, 26 January 2002, p 22). These sensors are far less intrusive, says Jorgensen.
But it remains to be seen whether the system can be improved enough to interpret normal speech. 鈥淪mall-vocabulary, isolated word recognition is a quite different problem from conversational speech, not just in scale but in kind,鈥� says Phil Green, a computer scientist focusing on speech and hearing at the University of Sheffield, UK. The preliminary tests may have been successful only because they were using short words, he says.
Not surprisingly, NASA says its main interest lies in space applications. The pressure in a spacesuit has to be kept low, otherwise the suit becomes stiff and restricts the astronaut鈥檚 movement. But if it is too low, speech becomes almost unintelligible. NASA is also keen to exploit the system in diving, where special masks are currently required for communication, and speech is affected by breathing gas mixtures containing helium.