A sample selective linear prediction analysis of speech

The formulation of linear prediction analysis using the generalized inverse matrix is reviewed and Givens´ reduction employing Gentleman´s algorithm is introduced as a fast computational method to obtain the least-squares estimate for the prediction coeficients. An improved version of the sample selective linear prediction (SSLP) is described and some performance examples on nonstationary synthetic speech are shown in comparison with those by the ordinary linear prediction analysis and the proto-type SSLP. Finally, the short-term SSLP is proposed and its discrimination performance on voiced plosives is compared with those by the ordinary short-term linear prediction analysis.