AT&T releases voice software

AT&T unveiled new voice software Tuesday, calling it "the most human-sounding computer-speech system in the world." The product is the first launched by AT&T Labs, and its ability to recreate any voice indicates intriguing possibilities for commercial applications.

The software, Natural Voices, is aimed at companies that have call-center operations, as well as at application service providers with voice portals and value-added resellers offering speech services. The voice created by the software is still a little too automatic sounding to compare to the soothingly sinister human-like tone of HAL in "2001: A Space Odyssey" but it's a step up for computer-synthesized speech, according to Bryant Parent, the product's manager.

With companies hoping to automate help-desk call-in operations as much as possible, real-sounding voices on the other end of the phone become more important, Parent said. "If the voice sounds terrible, callers will hang up," he said. For the moment, customers hoping to license AT&T's Natural Voices technology can choose from three voices - a male baritone named "Rich," a higher-pitched male voice named "Mike," and a female voice named "Crystal."

AT&T hopes to sell the software to business customers like the automotive industry, developers for cell phones, personal digital assistants and other electronics equipment makers looking to integrate voice technology into their products.

Natural Voices includes two parts, a software engine and a developer's kit. Customers can purchase the software engine and developer's kit immediately for $5,000. Licensing the software for commercial use in a call center costs $750 per port, plus 20 percent for each additional voice chosen.

Parent has a son who watches Barney - the talking purple dinosaur on TV - more than frequently, and Parent said it's getting on his nerves. Ironically, the software he's promoting could be his worst nightmare come true. Horror upon horror for parents, already reeling from kids hypnotized by the omnipresent children's television icon, technicians could take 10 to 40 hours of dialogue to create a baseline of phonemes and half-phonemes - the sounds and inflections making up spoken syllables in a word - to automate Barney's voice, which could then become even more pervasive than it already is.

Matched with computer-generated images like those developed for feature films, writers could crank out Barney-themed audio and video as fast as they can write.

Working for two years with animators and programmers, producers of the film "Final Fantasy: The Spirits Within" billed it as the first to employ synthetic actors. Though the movie opened to mixed reviews as critics pointed to a weak script, some also said that the visual texture of the computer-generated characters reached a new point in the development of photo-realistic digital imaging. For a moment or two, the screen can fool a viewer into thinking a scene is a live shot, even though nothing but bits and bytes made it onto the film.

Natural Voices isn't quite at the same level of duplicating human speech precisely. The voice still halts unnaturally from time to time or makes a slightly incorrect inflection on some words, but the gap is closing between human speech and computer speech.

The voices for AT&T's software draw from lengthy samples of real people speaking, so any voice can be automated using the same technique, Parent said. Companies with distinctive, highly recognizable spokespeople, like James Earl Jones for Verizon Communications or Jamie Lee Curtis for VoiceStream Wireless, could automate the celebrity voices for their call-center operations, he suggested.

Provided, of course, that companies would automate actors out of a job.

"This isn't going to replace people any time soon, and it's not meant to," said Michael Dickman, an AT&T spokesman. "But it's interesting that the technology is getting so close. It's getting hard to tell the real from the automated."

The IDG News Service is a Network World affiliate.