Australians have embraced speech technology using it to check frequent flyer points, undertake banking or book tickets, but customers still expect to speak to a female.

Whether recharging a phone account or ordering pizza, Phillip Tremain, ScanSoft (formerly SpeechWorks) user interface designer, said there is a strong preference to interact with a fun, happy, female voice, because it emulates call centres.

Surveys indicate that a female voice is perceived as service-orientated, he said, whereas a male voice commands more authority.

"If your recharging a phone you want a fast, snappy female voice but it is a very different tone when lodging a tax return," Tremain said.

"Here the ideal conversation partner is respectful, can show empathy and is responsive so we will get a male to ask questions such as what is your salary?

"Research has found people are more likely to tell the truth if they are speaking to a male."

Admitting sexism is alive and well in speech technology land, Tremain said there is a new 'man' set to debut at Telstra toward the end of this year.

He said Telstra has always used a female voice known as Karen, but the telco is launching a new SMS product that, for the first time, will feature a male voice which will be known as Lee.

Tremain's job is to design interactions with the caller by managing the entire interface including the feel and sound of the conversation.

"I manage the whole interaction between caller and machine and how you build that conversation and overlay the branding of the company," he said.

"A [human] call agent gets training to handle thousands of calls but when you have an automated consultant you want them to incorporate company-wide terminology which varies depending on whether you are an airline or a bank."

Voice selection begins with understanding who the callers are, establishing demographics and then going through an archetyping process to place callers into big groups and finding a voice that appeals to them.

"It is a big ask expecting people to talk to machines but Australians are embracing this technology which, within the decade, will be the first point of contact for every organization you call," Tremain said.

Customers are always going to say something unpredictable, but Tremain said ScanSoft has speech scientists that build grammar and ask questions in a way that directs callers, using a question such as, do you want A, B or C.

This limits variations and statistical probability is used to establish the intent of the caller.

Moreover, the technology is being continually tuned to the local accent as more systems are deployed.

"The technology accumulates utterances which it feeds back so it can recognise more words and performance improves," Tremain said.