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

[hts-users:00593] Re: Optimal mcep order


Hi, Nicholas,

> Signal processing is not my trade...
> I was wondering how mcep order relates to sample rate.
> HTS uses order 24 for 16kHz which seems to be fine.
> I'd suppose that other values might be better for
> HTS-based synthesis at 8000Hz and 22050Hz.
> Any comments on this?

For a lower sampling frequency, a lower order of mel-cepstral
analysis is better, and vice versa.  However, the optimum order
is not proportional to the sampling frequency.  I may recommend
12-16 for 8kHz sampling.

Also, we should adjust the frequency warping parameter alpha (-a
option of mcep) for each sampling frequency.  You may find
examples of alpha in Table 1 in the following document:

  http://www.sp.nitech.ac.jp/~tokuda/tips/mgceptr_sa2.pdf

We have observed that the optimum set of order and alpha is
speaker-dependent, unfortunately..

Keiichi Tokuda
tokuda@xxxxxxxxxxxx
http://www.sp.nitech.ac.jp/~tokuda/

References
[hts-users:00591] Optimal mcep order, Nicholas Volk