This paper deals with the problem of estimating and reducing residual echo components that result from reverberant components beyond the length of FFT block. The residual echo reduction process suppresses the residual echo by applying a multiplicative gain calculated from the estimated echo power spectrum. However, the estimated power spectrum reproduces only a fraction of the echo-path impulse response and so all the reverberant component are not considered. To address this problem we introduce a finite nonnegative convolution method by which each segment of echo-impulse response is convoluted with a received signal in a power spectral domain. With the proposed method, the power spectra of each segment of echo-impulse response are collectively estimated by solving the least-mean-squares problem between the microphone and the estimated-residual-echo power spectra. The performance of this method was demonstrated by simulation results in which speech distortions were decreased compared with the conventional method.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.