Selected applications of LPLE in speech processing

Linear prediction with low-frequency emphasis (LPLE), an all-pole modeling technique which emphasizes the lower frequency range of the input signal, was described by Alku and Backstrom (Alku, 2004). The method is based on first interpreting conventional linear predictive (LP) analyses of successive prediction orders with parallel structures using the concept of symmetric linear prediction. The experiments presented in this work are aimed to show that the LPLE method is well-suited for those speech processing and enhancement applications, where low-order all-pole models with improved modeling of the lowest formants are needed. This is done by replacing the LP block with a LPLE block. Distortion measures using Itakura distance and noise cancellation using a Wiener "type" filter were explored, utilizing the LPLE coefficients instead of LP coefficients.