Speech Parametrization and Recognition Using Block and Recursive Linear Prediction with Data Compression

Abstract

Results of experiments studying the use of LPC techniques in feature measurement and the posterior application of a fixed length frame compression technique in a isolated word recognition system are presented. The LPC analysis is performed by the classical block estimation method or by recursive estimation method which tries to make transient sound recognition more accurate. Frame compression technique compresses the stationary parts of speech signal and has several parameters to discern better the transitions.