Annoyance Model Driven Selective Active Noise Control

摘要

In this article, we present an annoyance model driven selective active noise control (ANC) system. Our approach uses Block Sparse Nonnegative Matrix Factorization (B-NMF) to identify the background noises that are present in the acoustic environment using any remote microphone recording, and assigns a soft membership score to each of the noises present. These scores have then been used along with an annoyance score computed for the noise types present in the environment, using a psycho acoustic annoyance model, to decide which are the most annoying sounds currently present in the environment. This information is further used in a Kalman Filter based noise canceler to just cancel the annoying sounds, but pass through everything else. This proposed selective ANC system delivers a pleasant yet acoustically aware listening experience to the user.