A combined vector and scalar codebook for robust quantization of LPC parameters

An important problem in speech coding is the quantization of linear predictive coefficients (LPC) with the smallestpossible number of bits while maintaining robustness to a large variety of speech material and transmission media. Sincedirect quantization of LPCs is known to be unsatisfactory, we consider this problem for an equivalent representation,namely, the line spectral frequencies (LSF). To achieve an acceptable level of distortion a scalar quantizer for LSFsrequires a 36 bit codebook. We derive a 30 bit two.quantizer scheme which achieves a performance equivalent tothis scalar quantizer. This equivalence is verified by tests on data taken from various types of filtered speech, speechcorrupted by noise and by a set of randomly generated LSFs. The two-quantizer format consists of both a vectorand a scalar quantizer such that for each input, the better quantizer is used. The vector quantizer is designed froma training set that reflects the joint density (for coding efficiency) and which ensures coverage (for robustness). Thescalar quantizer plays a pivotal role in dealing better with regions of the space that are sparsely covered by its vectorquantizer counterpart. A further reduction of 1 bit is obtained by formulating a new adaptation algorithm for the vectorquantizer and doing a dynamic programming search for both quantizers. The method of adaptation takes advantage ofthe ordering of the LSFs and imposes no overhead in memory requirements. Subjective tests in a speech coder revealthat the 29 bit scheme produces equivalent perceptual quality to that when the parameters are unquantized.