An adaptive post-filtering technique based on a least squares approach

Abstract

This paper presents an adaptive time-domain post-filtering technique based on the least squares approach and modified Yule-Walker (MYW) filter. Conventionally, post-filtering is derived from an original LPC spectrum. In general, this time-domain technique produces unpredictable spectral tilt that is hard to control by the modified LPC synthesis, inverse and high pass filtering and hence introduces muffling in the speech quality. Other approaches of designing post-filters were developed in the frequency domain which can only be used in sinusoidal based speech coders. We have also developed a new time-domain post-filtering technique which eliminates the problem of spectral tilt in the speech spectrum and can be applied to various speech coders. The new post-filter has a flat frequency response at the formant peaks of the speech spectrum. This post-filtering technique has been used in a 4 kb/s harmonic excitation linear predictive coder (HE-LPC) and subjective listening tests have indicated that this technique outperforms the conventional one in both one and two tandem connections.