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

[hts-users:02327] Re: question about the mcep in SPTK!!


You are right.

We are using an algorithm slightly different from that in the paper:
the algorithm used in the SPTK implementation updates
mel-cepstral coefficients c(m) directly
but the coefficients b(m) are updated in the paper.
This makes the difference.

Keiichi Tokuda

2009/11/12 박윤성 <ys.park@xxxxxxxxxxx>:
> Dear all HTS users,
>
>
>
> I’m analyzing the SPTK especially the mcep.
>
>
>
> But I have something confusing..!!
>
>
>
> In _mcep.c, therer are two frequency warping function freqt and frqtr.
>
>
>
> I think both functions are used for frequency warping. But I can’t find any
> relationship with the paper what I’ve ever seen.
>
>
>
> I saw the paper “An adaptive algorithm for mel-ceptral analysis of speech”
> written in 1992.
>
>
>
> I can’t find any relationship with the paper. What equation refer to the
> freqt and what equation represent the frqtr in the paper?
>
>
>
> Or did I see the wrong paper? (It was referenced in SPTKfre-3.2.pdf file)
>
>
>
> Best Regards,
>
> Youn-Sung Park
>
>