Design of bandwidth scalable LSF quantization using interframe and intraframe prediction

We developed a bandwidth scalable LSF (Line Spectral Frequency) quantizer utilizing interframe and intraframe prediction. The predictor used in the quantizer was designed as combination of first order AR (auto regressive) prediction and a codebook mapping technique, which maps a narrow band LSF codebook to a wide band LSF codebook. An upsampling process in the autocorrelation domain was utilized for conversion between narrowband LSF and wideband LSF. Codebooks for the mapping, coefficients for the predictor, and codebooks for VQ (vector quantization) of prediction residues were designed by using training database including multi languages and various kinds of background noise conditions. When 16 bits were assigned to an enhancement layer of the scalable LSF quantizer, utilization of the interframe and intraframe prediction improved the quantizer performance in spectral distortion by about 0.3 dB.