[hts-users:03480] Re: The synthesized waveform of HTS2.2(Straight) using LSP is silence

[那兴宇 <nxy-yzqs@xxxxxxx>]

It depends on the degree of instability.
If the LSP ordering property is broken, i.e. in some silent segments, it's hard for a global stabilization to act.

--

Xingyu Na (那兴宇)
http://www.idiap.ch/~xna/

Beijing Institute of Technology [naxy(at)bit.edu.cn]

Idiap Research Institute [xingyu.na(at)idiap.ch]

 

At 2012-11-26 21:27:47,ArthurLeo <bin007.zhao@xxxxxxxxxxx> wrote:

In the hts2.2 straight demo, there is the lspcheck function before  the lsp is transformed to the mgc.

If the  LSPs  predicted by the training model are unstable, Can lspcheck function make unstable LSPs  became stable?

> Date: Wed, 21 Nov 2012 17:14:25 +0900
> From: heigazen@xxxxxxxxxx
> Subject: [hts-users:03469] Re: The synthesized waveform of HTS2.2(Straight) using LSP is silence
> To: hts-users@xxxxxxxxxxxxxxx
>
> I expect
>
> 1. some generated LSPs became unstable
> 2. the MATLAB STRAIGHT code synthesized speech waveform in float,
> which became very large as filters were unstable
> 3. the MATLAB STRAIGHT code scaled the amplitude of speech waveform
> according to the maximum value while saving wav in short
> 4. as the maximum value was very large, almost all points in the wav
> file became 0
>
> Please check the generated LSPs are stable or not. You can convert
> LSP into spectrum and plot it using SPTK. Please refer to the SPTK
> manual.
>
> Regards,
>
> Heiga
>
>
>
> 2012/11/21 北语语音组成员李伟 <blculiwei@xxxxxxxxx>:
> > Hi, actually the matlab and straight are fine, because if they doesn't work,
> > I can't achieve synthesized wave using default settings.
> > BTW: I had checked the BAP file, it's not empty.
> > Thanks
> >
> > 2012/11/21 Thanh-Son PHAN <sonphan.hts@xxxxxxxxx>
> >>
> >> HT did not run, see B
> >
> >
> >
> >
> > --
> >
> > Li wei
> > Department of Information Science, Beijing Language and Culture University
> > Room 208a, Bulding 15, Beijing Language and Culture University, Haidian
> > District, Beijing,P. R. China
> > Post code:100083
> > Mobile: 15110081848
> > E-mail: blculiwei@xxxxxxxxx
> >
>
>
>
> --
> Heiga ZEN (in Japanese)
> Byung Ha CHUN (in Korean)
> <heigazen@xxxxxxxxxx>
>