[hts-users:00863] Re: about demo modification

["Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxx>]

Hi,

askhar@xxxxxxxxxx wrote (2007/10/19 11:14):

> I have found the problem and correct it, after that, I synthesized speech for
> my mother language.  I am glad in some extent, but the speech synthesized is just
> like a machine voice and not very clear, there are some noises.  If I want to
> improve the speech quality, should I adjust the configuration variable?  I adjusted
> the F0 range, it seems no use.  

Only F0 search range should be adjusted.

> I think I have to modify my context dependent
> attributes and question set, right? 

Yes.

> If so,how to make the modification?  According to the decision tree?

No, it should be modified according to the structure of your language.
For example, Japanese does not have tone but Mandarin has.
Context dependent formats of HTS-demo_Nitech-ATR503-M001 and HTS-demo_CMU-ARCTIC-SLT are designed for 
Japanese and English, respectively, according to their language structures.

> Is there any documentation about the hts system structure, for example, running
> flow, important file list and their function?
> If there is, could i have some copy of them?

No, we don't have any documentation and we don't have any plan to release it.

> I used only 10 mono, full, raw files, and put the same full lab files into the data/labels/gen directory.

It's too small, I guess this is the reason why you couldn't get good synthesized speech.
At least 50--100 utterances should be used to get intelligible voice.

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://www.sp.nitech.ac.jp/~zen
------------------------------------------------