Something worth looking at for VERY low bitrate transmission in the
reconstruction of speech from MFCC vectors.
People have shown that you can get the bitrate of MFCC vectors down to
about 800bps. I.e.
"Low bit-rate feature vector compression using transform coding and
non-uniform bit allocation
IEEE International Conference on Acoustics Speech and Signal Processing
(ICASSP)
B.P. Milner and X. Shao 2003"
It has also be shown that you can reconstruct the original speech quite
well from MFCC vectors.
"Clean speech reconstruction from noisy mel-frequency cepstral
coefficients using a sinusoidal model
IEEE International Conference on Acoustics Speech and Signal Processing
(ICASSP)
X. Shao and B.P. Milner 2003"
From:
http://www.cmp.uea.ac.uk/Research/speechgroup/publications.php
Cheers
Alastair