Ebert, however, would like broadcast quality, a tougher challenge that is spurring CereProc to consider a hybrid approach that uses the HTS model to select among stored phones, generating only less-common sounds that are missing or poorly represented in the database.
"It's great to have someone prominent like this [as a test case]—it moves the technology forward for us and makes it more obvious to other people that it can be done," Aylett says. His inner engineer has been piqued: "I just want to solve this problem." A small sample of the work in progress debuted on Oprah last year, but the date for a finished version is still uncertain.
The time it takes to type out speech will still hamper real-time conversation. Says Aylett, "It gives you real humility as an engineer when you realize that what you're competing against is a Post-It note."
A final question will only be answered when Ebert's new voice goes into service. Will it trigger "the uncanny valley," that is, the human revulsion to robots that are the wrong amount of almost human?
"I doubt if it will ever be a problem," Ebert says via e-mail, "but if it is, it's one I'd like to have."