[hts-users:00548] Re: help in SPTK(Now HMGenS)

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

Hi,

Mohammed Samir wrote:

> I follow your advice and HTS-2. has been installed
> (Thank you). Now I have 5 utterance of 1 phoneme, I
> have use HCopy to generate *.mfc files, then these
> files introduced with proto file to HInit to initiate
> HMM. Then I Used HRest with -g optin. Now I want to
> produce sound file(wave or any signal) from this
> output(  state duration pdf) can it be done using
> HMGenS only? and How?

You used HCopy to extract MFCCs and append dynamic features, didn't you?
If so, you should check "regression window coefficients" used in HCopy to calculate dynamic features.
HMGenS requires regression window coefficients which were used to calculate dynamic features in training stage.
And then run 

	$ HMGenS -C syn.cnf -H hmm.mmf -N dur.mmf gen1.lab gen2.lab ...

Please refer HTS-demo to specify options and configuration variables for HMGenS.


And I should mention that SPTK does not provide any tool to resynthesize waveforms from "MFCCs".
In HTS-demo, we use "mel-cepstral coefficients", not MFCC.
The difference between MFCCs and mel-cepstral coefficients has been discussed in this mailing list.
If you want to know the detail, please search the hts-users mailing list archive.

I saw several papers which attempted to resynthesize waveforms from MFCCs, but they are not provided in SPTK.
If you want to do it, please implement it by yourself.

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