[hts-users:00942] Re: Questions about certain concepts

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

Hi,

marc sobhy wrote (2007/11/18 20:14):

> the thesis used the mcp ? why he didn't use any other speech encoding 
> such LPC or MFCC ?

LPC coefficients have a stability problem if they are quantized excessively.
Similar problem may occur if they are modeled by HMMs.

Please see http://hts.sp.nitech.ac.jp/hts-users/spool/2006/msg00097.html why he didn't use MFCCs.

> when he tried to generate the cepstral coefficient, why he didn't used 
> the original F0 instead to regenerate it again with the spectrum

In their early papers, original F0s were used.
However, to realize text-to-speech synthesis, F0s should be generated 
because we cannot obtain original F0s for unknown sentences.

> Why he take the Log of the F0 when extract it from the speech databse

It worked better, and it is generally thought that human perception of frequency is logarithmic in nature.

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