Already, I have spent over sixty hours in a recording studio working with CereProc – to my mind the leading experts in the world on own-voice Text-to-Speech synthesis – pushing to the limit what is possible to voice-bank.
We’ve used two totally different technologies: Deep Neural Net (where we hope to build in expressiveness) and Unit Selection (unable to handle much emotion but still the best possible voice if you want to sound like you’re reading the News).
We finally finished all the necessary recording just before my voice became too bad to continue. But the important point is that my voice is now safe.
Now, with key members of the Think Tank, we’re going to devise an Artificial Intelligence (AI) system to drive my voice synthesiser, listening to what people are saying, and proposing suitable responses based on how I normally reply. At last, this offers Verbal Spontaneity to those of us that can only communicate using an eye-tracking system.
What’s more, one of the spin-offs from this research is that those of us that lose our voices are nevertheless going to be able to sing again! It’s very early days, and there’s still a load of research to complete, but here is a Proof-of-Concept of Peter 2.0 preparing for Christmas…