Telstra will start selling enterprise grade, home grown ‘natural language’ speech recognition engines directly into the Australian market over the next two years, according to its chief technology officer, Dr Hugh Bradlow.
Technical details, so far, are scant, but the technology will use user-voice executed commands combined with artificial intelligence systems to deliver services and transactions.
Delivering a technologist’s perspective of the Australian ICT future to the National Press Club last week, Bradlow said Australian enterprises with large customer bases would be looking for interactive voice technology that can recognise both phrases and context to deliver further savings.
“I’m very confident that if [enterprises] aren’t already looking at this sort of technology they will be over the next year or two as cost pressures build,” Bradlow said, citing the example of a customer being able to purchase multiple movie tickets at a specific time and place using nothing but voice commands.
He claimed speech recognition would also become “immensely powerful” as technologies able to recognise natural discourse were integrated with systems that were aware and enabled with individual customer or user contexts such as products or services.
Bradlow said that, to date, the Telstra Research Lab had also developed a commercially available software project called Lyrebird, although development costs were still commercial in confidence.
Asked when Telstra’s notorious automated directory assistance service would start to function without perpetual human assistance, Bradlow said it had suffered from a user education problem and was only ever intended to cater for about 300 key phrases.
Bradlow did not mince words in defending Telstra’s broadband performance.
“I dispute that we are falling behind,” he said. “It has taken Australia more than 100 years to achieve carrier-grade telephony network … there is no silver bullet.”
The current mix of cable, ADSL, ISDN and satellite was appropriate for a geographically dispersed population, Bradlow said.
While new interactive enterprise systems got the thumbs-up as the way of the future, he refused to throw his support behind audio-based biometrics as mainstream technology: “People can [over]hear passwords. [Then] you have to balance all the false positives and false negatives. You can put biometrics in place and then people will come along and break it. It will be a while before we can rely on biometrics as a consumer technology.”