A NON-PARAMETRIC BAYESIAN MODEL FOR SUPPRESSING THE INTERFERENCE IN THE ACOUSTIC ARRAY MEASUREMENT

Microphone array technology has a wide range of applications in areas such as mechanical noise source identification and aircraft noise source identification. The acoustic array is not only applied in anechoic chamber measurements, but also in in-situ measurements, which is an essential requirement for current applications. However, the microphone array measurements are inevitably affected by background interference when we wish to implement in-situ measurements. This paper proposes a general array denoising algorithm to address this issue. Considering the complexity of background interference, a Gaussian mixture model that can fit any probability distribution is constructed. The background interference tends to have non-independently and non-identically distributed characteristics between different microphone channels. The hierarchical Dirichlet process is applied to the Gaussian mixture model to avoid selecting the Gaussian component number. At the same time, a low-rank model of the sound source signal is constructed according to its correlation characteristics between microphones. All involved parameters of the proposed model are solved by the variational Bayesian inference. The sound source signal and the complex background interference are eventually separated. The performance of the proposed algorithm is evaluated in numerical simulation and laboratory experiments. Both the effectiveness and robustness of the proposed algorithm in suppressing the complex background interference are also verified.