A 2-CODEBOOK FORMAT FOR ROBUST QUANTIZATION OF LINE SPECTRAL FREQUENCIES

An important problem in speech coding is-the quantization of linear pr edictive coefficients (PC) with the smallest possible number of bits w hile maintaining robustness to a large variety of speech material and transmission media. Since direct quantization of LPC's is known to be unsatisfactory; we consider this problem for an equivalent representat ion, namely, the line spectral-frequencies (LSF), To achieve an accept able level of distortion a scalar quantizer for LSF's requires a 36 bi t codebook. We derive a 30 bit two-quantizer scheme which achieves, a performance equivalent to this scalar quantizer.;This equivalence is v erified by tests on data taken from various types of filtered speech, speech corrupted by noise and by a set of randomly generated LSF's, Th e two-quantizer format consists of both a vector and a scalar quantize r such that for each input, the better quantizer is used, The vector q uantizer is designed from a training set that reflects the joint densi ty (for coding efficiency) and which ensures coverage (for robustness) , The scalar quantizer plays a pivotal role in dealing better with reg ions of the space that are sparsely covered by its vector quantizer co unterpart, A further reduction of 1 bit is obtained by formulating a n ew adaptation algorithm for the vector quantizer and doing a dynamic p rogramming search for both quantizers. The method of adaptation takes advantage of the Ordering of the LSF's and imposes no overhead in memo ry requirements. The dynamic programming search is feasible due to the ordering property, Subjective tests in a speech coder reveal that the 29 bit scheme produces equivalent perceptual quality-to that when the parameters are unquantized.