[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:04298] Re: How to solve the clipped synthesis wav caused by using GV


Hi Qingsong,

If you scale the waveform during synthesis so that it doesn't clip,
does the resulting audio sound reasonable?  If so, depending on your
application, perhaps you just want to always use slightly quieter
synthesis (more headroom) to try to ensure it doesn't clip?

There is indeed a pathology in GV generation which can sometimes
introduce a loud pop, click or high-pitched whine artifacts into the
generated audio, and I discuss the causes of this quite a bit in my PhD
thesis
(http://mi.eng.cam.ac.uk/~sms46/papers/shannon2014probabilistic-thesis.pdf).
However typically the form of artifact introduced doesn't result in
clipping, just sounding bad.  It may be that the clipping you're
experiencing is due to the GV generation pathology or it may be
unrelated.

Cheers,

Matt


On Tue, 28 Jul 2015 08:25:50 +0100
Rasmus Dall <R.Dall@xxxxxxxxxxxx> wrote:

> I believe Matt Shannon's (Cambridge University) Thesis was about this
> issue.
> 
> Perhaps you could find something about it there?
> 
> - Rasmus
> 
> Quoting Qingsong Liu <liuqs.ustc@xxxxxxxxx> on Tue, 28 Jul 2015  
> 15:19:15 +0800:
> 
> > Hi, all,
> >
> >
> > It seems that the wav samples may be over-enhanced by GV and cause
> > overflow. the reason it is that GV change the values/range of
> > speech parameters.
> >
> > have anyone had these chlipped wave samples problem?
> >
> >
> > Qingsong
> >
> > --
> > ---------------------------------------------
> > Qingsong Liu
> > liuqs.ustc@xxxxxxxxx
> > Univ. of Sci.& Tech. of China
> > ----------------------------------------------
> >
> 
> 
> 


References
[hts-users:04296] How to solve the clipped synthesis wav caused by using GV, Qingsong Liu
[hts-users:04297] Re: How to solve the clipped synthesis wav caused by using GV, Rasmus Dall