Speech enhancement based on a simplified generalized sidelobe canceller structure

We present a beamspace realization of the generalized sidelobe canceller (GSC) applied to audio and speech capture, and show that it has some specific advantages compared to the standard GSC implementation for certain noise assumptions. Specifically, we demonstrate that for this application the herein proposed implementation is less prone to signal cancellation due to source positioning errors, and that it exhibits a better attenuation of low frequencies when broadband audio signals are captured by uniform linear arrays. Results are shown both for simulations with small uniform linear arrays of five and eleven elements, as well as for actual data recorded by a 300 element microphone array at the Staples Center sports arena.