Dr Andrew Hunt, senior consultant at speech technology vendor SpeechWorks International, said Karenâ€"the Australian voice created for its Speechify text-to-speech softwareâ€"took a lot of stamina from the person whose voice they recorded.
Hunt said the aim had been to find a voice that would be broadly useful to Australian telephone applicationsâ€"one which people would feel comfortable interacting and listening to.
"The most important thing was to choose a person with a voice that would represent organisations effectively," he said. "We didn't go for a prim and proper voiceâ€"we wanted someone who sounded like a general Australian speaker".
"[It's] important to have voice quality that's really consistent and carries the same tone," he said.
Once the voice recording for the product--of a woman called Karen--had been carried out, the basic units of soundâ€"phonemesâ€"were analysed. "A lot of the quality and naturalness in the final product comes from careful handling and management of those hundreds of thousand bits of sound," Hunt said.
The idea with text-to-speech systems is to provide a synthetic voice that becomes almost indistinguishable from human speech, according to a SpeechWorks whitepaper.
Examples of applications where text-to-speech can be used includes account or order information needs to be read out over the phone, locator services, and unified messaging.
6%
1%
