A multiframe parametric Wiener filter for acoustic echo suppression

Acoustic echo arises due to the acoustic coupling between the loudspeaker and the microphone in a full-duplex voice communication device. How to reduce or eliminate echo has been an important problem in voice communications. This paper deals with this problem in the short-time Fourier transform (STFT) domain. An approach to acoustic echo suppression (AES) is developed, which uses a linear filter in every STFT subband. A non-parametric variable step-size (NPVSS) normalized-least-mean-square (NLMS) algorithm is used to estimate the echo signal in every STFT bin and time frame. We then define the desired output signal as the sum of the desired near-end speech and a residual echo component. A parametric Wiener filter (PWF) is subsequently derived by minimizing the near-end speech distortion with a constraint on the residual echo. There are two parameters in this PWF, which can be adjusted to control the compromise between the amount of echo attenuation and the degree of near-end speech distortion, and the level of the residual echo remaining at the output. These two parameters can be tuned to optimize the perceptual quality of the echo-reduced signal. Experimental results demonstrate that the developed PWF can effectively attenuate the undesired echo and estimate the near-end signal with

little distortion.