Evaluation of improved articulatory codebooks and codebook access distance measures

An improved method for generating articulatory codebooks is introduced. The distance measure used for codebook access is optimized. The improvements are evaluated and compared in terms of a perceptually motivated formant distance measure. It is found that the codebooks derived from the articulatory model of P. Mermelstein (1973) are more robust against model/speaker mismatches than the codebooks derived from the model of C. Coker (1976).