For anyone who has ever had problems with voice recognition software on phone menus or elsewhere, senior Intel executive David Perlmutter can relate. He says voice and handwriting recognition are among the bigger obstacles to people working more naturally with computers.
"Typing is not necessarily natural," Perlmutter said during an interview at the Intel Developer Forum in Beijing. Handwriting and voice recognition will make it much easier for people to use computers, he added, but "this seems to be a tougher problem to solve, beyond more capable computing, it's just the algorithms are not good enough."
For voice software, "the real test is recognizing my [Israeli] accent, because I'm talking to all these things on the telephone and I go back to touching," he said.
Work on voice recognition software has been going on for years and a number of applications exist for speech-to-text programs on computers, automated phone systems for banking and airline ticketing, and voice commands for mobile phones and cars.
Voice-recognition technology developer Nuance, for example, sells Dragon NaturallySpeaking voice-to-text software that learns to recognize your voice, while Microsoft has worked with Ford on Sync, a speech recognition software in some cars that allows people to ask for directions, search for music, answer phone calls and more while driving.
Despite years of work, typing is still the main way to input data on most devices.
Perlmutter, co-general manager of the Intel Architecture group and seen as a potential successor for company CEO, believes the human-to-machine interface, beyond voice, will improve in coming years. The world is already moving from text to mouse to touchscreen and "we'll continue to move into more gesture recognition," he said.
"There's a lot of understanding of what you want from gestures, like the way human beings communicate, not just listen to the words, to see also gestures, hand movements, body movements, face," he said. "That's going to get better."
Machine-to-machine interface is also going to get better, "which is really important, so you find out that machines are interacting," Perlmutter said.
But all that still leaves the challenge of voice recognition.
"Maybe in 20 years it's going to be resolved," he said. "I'm still waiting."