[hts-users:00303] Re: training error - no proto model
- Subject: [hts-users:00303] Re: training error - no proto model
- From: "Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxxx>
- Date: Thu, 18 May 2006 06:37:20 +0900
Hi
donall.mccann5@xxxxxxxxxxx wrote:
If the program is searching for a model with the same center phoneme, could
it be a problem with the amount of training data (only 30 sentences) or am
I thinking along the wrong lines entirely?
You said that you are using 30 sentences, but we have successfully constructed an hts voice using only 5 sentences which
were designed to include all phonemes (of course its quality was not good).
So in my opinion it is not a problem of HTK, it is problem of design of your sentences.
Currently, there is no way to avoid this problem except including all phonemes in the training data on the current
implementation or substituting unseen phonemes using the other ones.
If you have any idea to solve this problem, please report to this mailing list.
Regards,
Heiga Zen (Byung Ha Chun)
--
------------------------------------------------
Heiga ZEN (in Japanese pronunciation)
Byung Ha CHUN (in Korean pronunciation)
Department of Computer Science and Engineering
Nagoya Institute of Technology
Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan
http://kt-lab.ics.nitech.ac.jp/~zen
------------------------------------------------
- References
-
- [hts-users:00302] Re: training error - no proto model, donall . mccann5