Long Short-Term Memory and Convolutional Neural Networks for Active Noise Control

Abstract

Active noise control, or adaptive noise cancellation, techniques (ANC) attempt to reduce sound pressure level by generation and superposition of an anti-noise signal in order to improve human safety or comfort in noisy environments. The least mean squares (LMS) algorithm and variants as well as some architectures of neural networks have been employed successfully for active control of noise. This work presents the novel use of long short-term memory (LSTM) and convolutional neural network (CNN) architectures for this task. Testing on a selection of commonly used noises, improved results are demonstrated when compared to both the traditional LMS approach and previously published neural-network approaches.