Non-verbal utterances could one day be the way to control personal computers and navigate through call-centre systems, reckons Takeo Igarashi of Brown University in Providence, Rhode Island.
Instead of having to speak carefully and deliberately to make a voice recognition system understand you, Igarashi wants to use what he says are the much more distinctive properties of pitch and duration in our 鈥渉uhs鈥�, 鈥渕mms鈥� and 鈥渦h-ohs鈥�.
Voice recognition programs typically measure the frequency variation within an utterance, carving it up into words, and making a stab at the meaning. But all too often, these systems fail or take ages to respond, says Igarashi: 鈥淪o we tried to see what we could do without words.鈥�
Advertisement
The system he鈥檒l describe at a computing conference in Orlando, Florida, next month measures the pitch and duration of quieter grunt-like sounds like 鈥渁h鈥� and 鈥渦mm鈥� that many people make while they鈥檙e working.
Igarashi thinks his system will work particularly well for continuous interaction with a desktop computer. It would be ideal for scrolling through a document or map on the Internet, or skipping through tracks on a CD player (see Graphic). A quick 鈥渦h,oh鈥�, for instance, could produce an instant Undo on a PC 鈥� reversing an action far faster than you can use a mouse, Igarashi says.
鈥淕runt shorthand鈥� will allow developers to choose a set of sounds that are easily distinguishable by a computer, comments Tim Hazen of MIT鈥檚 Spoken Language Systems Group. But he wonders whether people working in crowded offices would be happy to grunt away in front of their colleagues.