Fast low bit rate lattice entropy coding for speech and audio coding

In this work we study variousmethods for low bit rate encoding of long subvectors of transform coefficients from speech and audio codecs. Given the data characteristics, we propose a lattice based method that takes better into account the sparsity of the signal. We also propose two generic lattice encoding techniques addressing the encoding efficiency when encoding several lattice codevectors together and computational complexity of the lattice codebook nearest neighbor search. We examine as well the effect of different core codecs for speech and audio on the differential data characteristics and how the input signal type influences the coding efficiency of the considered methods.