A Beam Shape Definable and Low Computing Cost Adaptive Audio Beamformer

Audio beamformer utilizes a microphone array to pick up sounds from a designated direction while suppressing sounds from other directions as much as possible. It has a wide range of applications in areas such as anti-interference recording, zone recording, private calls, and audio zooming. The adaptive beamformer can automatically adjust its coefficients according to the spatial distribution of the signal, thus having much better performance than the fixed beamformer. This paper proposes a beam shape definable and low computing cost adaptive audio beamformer, which first uses the generalized cross-correlation (GCC) method to calculate the spatial spectrum distribution of the audio, and then uses this spatial spectrum distribution information to guide the generalized eigenvalue decomposition (GEVD) based method to perform beamforming with maximum signal-to-distortion ratio (SDR). In addition to definable beam shape and low computing cost, the proposed method also has the characteristics of fast convergence (good real-time performance), insensitivity to array shape, and no nonlinear distortion. This paper experimentally verified the algorithm’s ability to perform directional speech pickup and speech separation. In the simulation test, the proposed method can improve the SDR of the speech signal in the target direction by 11.33 dB, and in the real-device test, it achieves the same level of results. It also performs well in noisy and reverberant conditions. We have provided samples of processing results for the readers’ listening experience.