UNSUPERVISED NOISE-AWARE ADAPTIVE FEEDBACK CANCELLATION FOR HEARING AID DEVICES UNDER NOISY SPEECH FRAMEWORK

This paper proposes a novel adaptive feedback cancellation (AFC) architecture to improve the performance of an existing robust AFC method in the presence of noisy speech conditions. By employing a computationally efficient Spectral flux (SF) feature-based unsupervised voice activity detector (VAD), we adaptively control the step sizes in the proposed AFC algorithm (SFPEM-AFC). The proposed AFC method achieves faster convergence and lower misalignment errors than earlier methods. Objective evaluation of the AFC algorithm is presented using Signal to Feedback Ratio (SFR) and Misalignment (MISA) values for several noisy conditions. The Proposed method shows a significant reduction in the MISA values while maintaining higher SFR and higher perceptual quality over the earlier methods. Experimental results are presented for realistic noisy conditions to demonstrate the superiority of the proposed noise-adaptive AFC method for hearing aid devices (HADs).