Perceptual quality of hybrid echo canceler/suppressor

Acoustic echoes arise whenever there is acoustic coupling between a loudspeaker and a microphone. A traditional solution for eliminating the undesired echo signal is an acoustic echo canceler (AEC), which identifies the echo path between a loudspeaker and a microphone by means of an adaptive filter. The echo signal can be canceled successfully when the modeling filter approaches the true echo path. In practice, however, a modeling filter often differs from the true echo path due to complicated reasons such as environment changes, the lack of knowledge about the length of the echo path, and so on, resulting in residual echo signals. Another way to mitigate the echo effect is through echo suppression. Unlike an AEC, an acoustic echo suppressor (AES) achieves echo attenuation by means of spectral modification. This approach usually has a much lower complexity, and is robust against minor echo path changes. However, it sometimes introduces audible distortions to the processed signal. Many practical systems combine an AEC and an AES in a sequential way such that the former achieves major echo cancellation, and the latter attenuates the residual echoes. In this paper, we propose a novel hybrid approach for addressing the echo problem. The full-band signal is split into two frequency bands, with AEC being applied to the lower band, and an AES to the upper band, respectively. Through subjective tests we demonstrate the superior quality and robustness of this new method, compared to a full-band AEC or a full-band AES.