Closed jurihock closed 2 years ago
Actually, the discrete cosine transform of x[n]
is the real part of its Fourier transform X[m]
. That's what the Cepster module computes.
The [2] shows an additional trick to half the packing procedure, as implemented here. This trick doesn't seem directly applicable in [3] to me, due to pre-calculated coefficients A
and B
.
So I currently don't expect procedure [2] to be much more beneficial or faster than [3]... 🤔
The forward and backward transformations of the log magnitude spectrum are both purely real. Thus think about an optimization possibility...
See also: