Computer-implemented method for generating anti-noise

Abstract

A method for generating anti-noise comprises receiving a sound signal representative of ambient sound including noise from a noise source, anti-noise from an anti-noise generator, and propagation noise from environment; processing the sound signal using a deep learning algorithm configured to generate an anti-noise signal to form anti-noise; and outputting the anti-noise signal to the anti-noise generator. The deep learning algorithm features an iterative encoder module forming plural feature maps; an attention module generating plural attention maps respectively based on the feature maps; a recurrent neural network (RNN), with long short-term memory layers receiving the feature map of the final iteration of the encoder module, predicting a future portion of the sound signal and modelling temporal features of the feature map of the final encoder module iteration; and an iterative decoder module mapping the output of the RNN to the anti-noise signal having common dimensions as the sound signal.