From: Jessica P. H. <jph...@ar...> - 2002-09-25 21:57:14
|
On Wed, 25 Sep 2002, robert b wrote: > I think the issue of training via a particular piece of text is a much smaller problem than > getting the adaptation algorithms (i.e., mixing the speaker-independent model data with a > particular speaker's data) to work and other training issues. And I don't think a model generated off of a computer-generated voice would be a lot of good for recognizing normal human speech. As Robert said later in his mail, I think good speaker-independent models are generated using data from a lot of different people. It was an interesting suggestion, though :) j -- "Users complain that they receive too much spam, while spammers protest messages are legal." -InfoWorld "You do not have to do everything disagreeable that you have a right to do." -Judith Martin (Miss Manners) |