Comparison of several clustering algorithms for data rate compression of LPC parameters

The purpose of this research was to investigate the effectiveness of several clustering algorithms for separation of LPC speech segments. Four thousand frames of speech parameters were divided into thirty clusters using four clustering algorithms and three initial seed point selection methods. A second-order and an eithth-order norm in the area function parametric domain were used for distance measures. For single-pass clustering algorithms, using every 130th frame as a seed point resulted in a substantially lower error than that given by using the first thirty frames as seed points. The benefit was noticable for using a more complex algorithm for generating the initial seed points. For iterative clustering algorithms, the initial allocation had negligible effect on the final error and on the number of iterations.