The University of Western Australia (UWA) has received a $230,000 Government grant to perfect technology that will enable speech recognition in noisy environments.
The technology could provide voice-activated controls in settings where users cannot always be right next to a microphone.
"Users have most difficulties with [current] speech recognition [technology] when using it in uncontrolled environments," said Roberto Togneri, an Associate Professor at the university who is leading the project.
"We're trying to look at ways to reduce errors, for example through front-end processing," he told iTnews.
Speech recognition has become increasingly pervasive in the industry, with organisations like National Australia Bank and Centrelink deploying biometric voice systems that allow callers to verify their identity by speaking into the phone.
The technology is also gaining popularity among consumers, with both Telstra and Optus chiefs highlighting speech-to-text at the Mobile World Congress in Barcelona last year.
But the technology is not yet perfect, Togneri said, explaining that current systems require users to speak directly into a microphone in a relatively quiet environment.
"The key feature that makes speech recognition work is also its Achilles' heel," he said, explaining that the current pattern-matching approach tends to produce errors.
Togneri suggested that better front-end processing algorithms could have a more accurate result.
He said there could be "some interest" in commercialising the technology, depending on the direction of the research -- which depends on "who we work with in future".
Once developed, the technology could be used to assist the elderly and disabled, enhance Australia's defence capabilities through automated surveillance, and improve inventory technology in the local WA animal husbandry industry, he said.
