Kennedy, "Language identification using shifted delta cepstra
," in Proceedings of the 45th Midwest Symposium on Circuits and Systems, vol.
This algorithm extracts and concatenates the two kinds of feature vectors (long-term temporal cepstra
and harmonic structure information), which form the 26-dimensional feature vectors by using FFT, DCT, and IDCT in sequence .
Then, the harmonic structure information [q.sub.t](i) is obtained by lifting out the lower and higher cepstra
because the medium cepstra
capture the harmonic structure Information: