Robust sound source localization in reverberant environments using sparse Bayesian learning with dual regularization

Sound source localization in enclosed environments is strongly degraded by multipath reflections and background noise, which introduce high spatial correlation and reduce the robustness of conventional localization algorithms. To address this challenge, this paper proposes a dual-regularized sparse Bayesian learning approach for robust sound source localization in reverberant rooms. A frequency-domain propagation dictionary constructed from spatial impulse responses is used to characterize the acoustic transfer between candidate source locations and the microphone array. Within this framework, a dual-regularization strategy combining ℓ₁ and ℓ₂ penalties is introduced to construct a stable sparse prior that promotes source sparsity while reducing the influence of correlated reflection components. The proposed method is designed to improve localization robustness under low SNR ratio and strong reverberation conditions. Hyperparameters are estimated within the Bayesian inference procedure to enhance model stability and reduce sensitivity to noise variance. Simulation and experimental results demonstrate that the proposed method improves localization accuracy while achieving more effective sidelobe suppression. The proposed approach provides a practical and robust solution for sound source localization in real reverberant environments.