Active attenuation of free field noise in a spatial domain using neural networks inferred controller

Abstract

In this talk, we explore the application of deep neural networks (DNNs) to infer the decoupling of an incoming acoustic array, distinguishing between the signal of interest and noise components. The objective is to leverage this decoupling information to develop an effective controller that can attenuate the noise, thereby improving the quality of the desired signal. The decoupled noise is then fed into a specially designed controller to attenuate the noise while preserving the desired signal. The controller leverages the DNN's predictions to adaptively adjust its parameters, allowing it to adapt to changing noise conditions and improve noise reduction performance. The approach is evaluated using simulations and real-world acoustic array measurements. The results demonstrate the performance of the DNN-based decoupling and noise attenuation system in reducing noise while preserving the quality of the desired signal. The proposed system offers effective adaptive solutions for enhancing the quality of desired signals in the presence of noise, with potential applications in areas such as speech recognition, audio communication systems, and environmental noise suppression.