[hts-users:01463] Re: hts_engine-0.99 does not generate normal speech in HTS-demo using 12-th order LSP

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

Hi,

Xuchen Yao wrote (2008/06/25 12:17):

> I noticed that many speeches generated by hts_engine are truncated while 
> that generated by SPTK are normal. Here's my running environment:
>
> configs:
>
> Config 1: 12-th order LSP,     linear gain
>
>                  speeches generated by SPTK are OK, while wave files 
> generated by hts_engine are truncated (silence in the wave files).

Oops, this is because hts_engine API doesn't support linear gain.
I discussed it with Keiichiro Oura, maintainer of hts_engine API.
We will support linear gain in hts_engine API 1.0.

> Config 2: 24-th order Mel-cepstrum
>
>                  speeches generated by SPTK and hts_engine are both OK.
>
> 2. Speeches in Config 2 sound better than those in Config 1. Could 
> someone provide any experience that on what condition the speech 
> generated from LSP features achieve the best quality?

From our internal experiments we found that frequency warping strongly 
affected the final speech quality.
So how about trying to use mel-LSP rather than LSP?

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
------------------------------------------------