Diffusion-Based Distributed Multi-Frame Kalman Filtering With Speech Distortionless Constraint for Speech Enhancement
Qingying Zhao, Ruijiang Chang, Zhe Chen, Fuliang Yin · IEEE Transactions on Audio Speech and Language Processing · 2025
The widespread adoption and interconnection of intelligent devices equipped with microphones has further propelled the development of speech enhancement techniques in wireless acoustic sensor networks (WASNs). To further adapt flexibly to different environments, a diffusion-based distributed multi-frame Kalman filtering method with speech distortionless constraint for speech enhancement is proposed in this paper. Firstly, a multi-frame Kalman filtering method with speech distortionless constraint is proposed to suppress non-stationary noise and reduce speech distortion in each node. Then, to improve the transmission efficiency and alleviate the computational burden, the diffusion strategy is adopted to implement distributed multi-frame Kalman filtering through local communication among nodes to enhance the speech per node collaboratively. Finally, an adaptive diffusion weight estimation method based on the maximum likelihood criterion is proposed to enhance the robustness of the algorithm against abnormal situations. The proposed method can effectively suppress noise and improve speech quality in different noisy scenarios. Furthermore, it requires communication only among the microphones of the local node and neighbor nodes, thereby reducing the communication load. Experimental results validate the feasibility of the proposed method.