Fenris Wulf wrote on Sun, 12 December 2010 07:20 |
And has anyone noticed that digital speech synthesis doesn't sound any more natural than it did 25 years ago?
|
This is totally not true.
Check out Phonetic Arts and some of their tech-demo. Their runtime library can be trained to sound ridiculously well. Google just bought them, so I guess we picked the right system for our text-to-speech work since we were not the only ones thinking Phonetic Arts does it best.
The runtime can have the voice go from a slight whisper, to yelling in an angry manner in real-time during one and the same sentence. I use the offline (not runtime) stuff myself at work though, it doesn't sound nearly as convincing but it's only for autogenerating placeholders anyway.
There is a tech demo where they trained the system with Bill Clintons voice, and it sounds like Bill Clinton telling you where to go and what to do.
I figure a lot of time when synthetic speech is being used, people actually do want it to sound like synthetic speech and therefore use ancient stuff.
Cheers,
Danko