L1-regularised MV beamformer in the spherical harmonic domain for spherical microphone array recording

Teleconference systems are widely used in business and education. They are required to enhance only the target speech while suppressing diffuse or interference noise. The minimum variance beamformer is a strong method to enhance the target speech; However, it is less robust at low frequencies. We propose a robust beamformer based on the minimum variance method in the spherical harmonic domain. Less robustness is caused by the use of high-order spherical harmonic coefficients at low frequencies. To avoid the use of high-order coefficients in the filter calculation, we apply L1-regularization to the minimum variance method. Our analysis demonstrates the tradeoff between the robustness and spatial resolution and how the L1-regularization improves the robustness at low frequencies.