Probabilistic Spatial Filter Estimation for Multi-Channel Signal Enhancement in Hearing Aids

Abstract:
We propose a method to estimate the full noise covariance matrix in a multi-microphone signal enhancement setting from estimated target source presence probabilities. The latter are obtained from a previously developed probabilistic source localization algorithm. A-priori knowledge of temporal segments that contain noise only, thus is not required. An estimate of the target source-related steering vector is obtained, as proposed in previous work [1]. A standard MVDR system is employed for subsequent signal enhancement. Evaluation is carried out for anechoic and reverberant (office) conditions using 6-channel hearing-aid input signals. Results indicate that the proposed method outperforms an anechoic, isotropic-noise model when a-priori information is unavailable: I.e., in (a) anechoic condition with localized interferer in addition to isotropic noise, and (b) reverberant office condition. In these conditions, the proposed method and constrained versions thereof improve upon an entirely model-based approach by up to 16.1 dB and 7.7 dB SINR, respectively.