How do you talk to someone without opening your mouth? Psychics call it telepathy. NASA refers to it as subvocal speech.
Scientists at the NASA Ames Research Center in California have developed a system of tiny sensors that read nerve signals in the throat that control speech. You may not make a sound when, say, you read silently, but your nervous system is buzzing with activity. Recently, they used the system to make the first subvocal cell phone call.
The “silent speech” research was launched to deal with the high-noise environments encountered by astronauts. Of course, space itself is silent. But according to NASA scientist Chuck Jorgensen, there are many factors, from machinery to pressure changes, that can make communication difficult for astronauts onboard the shuttle or sporting a space suit. Related, he adds, is the drive to develop alternative human-machine interfaces such as speech recognition.
Speech recognition is only practical if the computer can hear what its operator is saying. Earlier, Jorgensen and his colleagues demonstrated sensors that when applied to your hand would detect subtle muscle signals known as electromyogram (EMG) signals. That EMG data could then be amplified to control a computer or robot appendage. The question that arose was whether a similar approach could be used to solve the noise problem surrounding vocal communication and speech recognition?
“We realized that it might be possible to intercept the signals that the brain sends to the vocal system, extract them and understand them before the person produces a sound,” Jorgensen explains.
Last year, Jorgensen and collaborator Brad Betts showed off their silent speech prototype system for the first time. The user wears button-sized sensors under his chin and near the Adam’s apple. Those sensors measure the nerve signals that control the vocal chords, muscles, and position of the tongue.
“It’s very much like whispering or reading to yourself,” Jorgensen says. “You don’t have to move your mouth though. In one demonstration, I’m communicating while holding my lips closed.”
The electrical nerve signals are delivered from the sensors to an amplifier and a digital signal processor that filters out the noise. Finally, software scours the data to identify the signature patterns corresponding to words that the system was programmed to recognize. When the silent speech technology was unveiled, the software was trained to identify six words such as “stop,” “go,” “alpha,” and “omega,” and ten digits. Since then, they’ve increased its vocabulary to around twenty words and begun teaching the system to detect the telltale EMG signals of vowels and consonants, the first step in building a full-blown speech recognition system.
By David Pescovitz