Modeling late reverberation in acoustic echo suppression

The traditional approach to cope with acoustic echoes is the use of an acoustic echo canceler (AEC). Usually, subband acoustic echo suppression (AES) is used after the AEC to suppress residual echoes which have not been subtracted. Recently, the use of only AES has been proposed without precise echo path estimation, but modeling parametrically the echo path with a delay and a single real-valued gain at each frequency of the short-time spectra. This approach captures only a fraction of the length of a typical echo path. We propose an extension to this simple echo path model, based on the consideration and estimation of the exponential decay of room impulse responses. Simulation results show an improvement of near-end speech quality when including the proposed reverberation modeling in AES. An important feature of the proposed AES system is its robustness against phase distortions, as only magnitude information of the loudspeaker and microphone signals are considered.