Experts set to create technology that trips off the tongue

Scientists from the University of Edinburgh hope the mathematical and computer systems will be able to understand complex speech and mimic language.

Current voice recognition programmes struggle when more than one person is speaking or in noisy environments and require an individual to speak directly into microphones.

But the five-year, 6.2 million project in collaboration with the universities of Cambridge and Sheffield aims to be able to distinguish voices in a "natural" way, including varied accents across the English language.

Professor Steve Renals, of the University of Edinburgh's School of Informatics, who is leading the project, said: "We are mainly used to speech recognition for call centres or smart phones or desktop dictation. So we want to improve the quality of those as a large part of our work.

"But you could also be able to control things in your home by speech, and particularly as people get older and less mobile, it can be good to control things just by speaking.

"If you could have some control of domestic devices, it could improve the quality of people's lives and allow them to stay in their own homes for longer."

He added: "Speech is something you do effortlessly as a person, but for technology it's quite difficult."

In human conversation, voices overlap and react spontaneously, said Prof Renals, presenting difficult challenges for speech recognition programs. The scientists will try to build models that could allow future software to record business meetings or control the family home.

Prof Renals said it would use a degree of artificial intelligence, though not in the sci-fi sense. The computer could take signals from speech and interpret them correctly.

He said: "You will never have a perfect speech recognition machine. We are trying to reduce the errors systems make.

"One of the things we are looking at is the great variety across Scottish-English and we would like to be able to recognise accents without training (the programs] which would be a huge task.

"There have been advances that lead us to believe we have a reasonable chance of progress."

The scientists said the technology could also help develop web search engines for audio clips, and become familiar with an individual's vocabulary, accent and vocal expressions.

Dr Thomas Hain, of the University of Sheffield's Department of Computer Science, said: "Speech technology is clearly on its way into the mainstream, but the key to its success is human-like performance.

"We are working on the foundations of technology that will bring speech to new users and exciting new applications from schools to board rooms."