Recovering the quality of speech degraded by reverberations in a room

Abstract

Speech signals captured by distant microphones in a room are usually reverberated due to wall reflections. Reverberation may seriously deteriorate the signal characteristics, thus damaging the quality of such applications as hands‐free telephony or automatic speech recognition. To eliminate such room reverberation effects, the inverse‐filtering of the room transfer functions (RTFs) between a speaker and the microphones appears to be a very promising approach. When applied to a microphone system, this processing should work blindly because no explicit information on RTFs or source signals is available. The blind design of an inverse‐filter set may be roughly classified into three approaches: (1) calculating an inverse‐filter set of RTF estimates obtained from observed reverberant signals based on the ‘‘subspace method,’’ (2) calculating the filter set from observed signals and replicas of direct‐sound signals using statistical properties inherent in speech signals, and (3) calculating the filter set direct...