Binaural Model Based Adaptive Binaural Noise Reduction

This paper presents an efficient binaural noise reduction scheme built upon the integration of adaptive noise cancellation technique with the perceptual binaural model. In particular, an intermittent ANC is utilized to cancel the noise during the speech pauses detected by a simplified binaural model, which shows an improved VAD detection performance especially in low SNR situations. Delayless subband processing is also incorporated, making it possible to apply ANC only in selected subbands and leading to reduced computational complexity accordingly. Simulations show that our proposed strategy can effectively reduce speech-shaped binaural noises in an efficient way.