Parallel adaptive bone-air conduction speech fusion with momentum-based narrowband noise cancellation for industrial environments

Abstract

Adaptive bone-air conduction speech fusion systems offer an effective solution for low signal-to-noise ratio (SNR) scenarios. However, sharp noises from rotating and cutting machines hinder complete narrowband noise elimination, thereby reducing speech intelligibility. To address this issue, this paper proposes a fully parallel adaptive bone-air conduction speech fusion system with integrated simultaneous narrowband noise cancellation. Non-acoustic sensors capture the dominant narrowband noise frequency components of environmental noise, while a momentum-adaptive algorithm constrained by a judgment factor enhances tracking robustness and suppression efficiency of narrowband interference in both air-conducted (AC) and fused speech signals, resulting in a purer residual error. Extensive simulations and experimental evaluations demonstrate that the proposed system effectively restores speech corrupted by narrowband noise, exhibiting superior tracking accuracy, reduced residual error, and minimized residual noise fluctuations in various industrial environments dominated by strong narrowband noise sources.