About time-variant filtering of speech signals with time-frequency distributions for hands-free telephone systems

Joint time-frequency distributions represent the energy or intensity of signal simultaneously in time and frequency. In this paper we introduce a smoothed version of the method denoted as S-method (SM) [18]. It is shown that it is an effective tool for representation of speech signals disturbed by noise. As a consequence, it can be efficiently applied to time-variant filter problems, as they occur in hands-free telephones. The optimal time-variant filter is derived in terms of the smoothed SM, and is illustrated by examples.