Systolic codebook generation (speech recognition)

Two systolic codebook generators which are based on finding clusters in a multidimensional space with a suitable dissimilarity measure are presented. They are implemented on two novel VLSI architectures. For segments of speech represented by a sequence of k linear-predictive-coding vectors, the first version finds r large clusters in time O(rk), and the second version finds the maximal r clusters in time O(r log k).<
>