[hts-users:00675] Re: LPC coefficient for syntheis

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

Hi,

Fang Li wrote (2007/05/17 16:59):

> I just leaned that LSP synthesis showed better performance than MCEP synthesis.
> But I found it sounds a little noisy.

Did you use LSP (linear) or mel-LSP (frequency-warped) ?
In our experiments we found that mel-LSP is better than mel-ceptrum but LSP (linear) is worse.
You can extract mel-LSP using SPTK.

> So now I want to use LPC coefficient for both training and synthesis in HTS-2.0.

I don't recommend you to use HTS with LPC because quantization and interpolation properties of LPC are not good.
As you know, modeling and synthesis processes are closely related to quantization and interpolation, respectively.
I've never tried, but I guess HTS doesn't work well with LPC.

On the other hand, LSP has good quantization/interpolation properties.

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