Deep learning-based active noise control for airplane cockpit

基于深度学习的飞机驾驶舱主动噪声控制

• 批准号: 533690-2018
• 负责人: Cha, YoungJin
• 金额: $ 1.82万
• 依托单位: University of Manitoba
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2018
• 资助国家: 加拿大
• 起止时间: 2018-01-01 至 2019-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=647018
• 关键词: Deep; learning; based; active; noise

Vintec Acoustics focuses on providing solutions to various engineering problems related to acoustics, including high-performance sound equipment design and control of noise and vibration. Industrial fields such as construction sites, automobiles or aircraft generate continuous noise, and this noise needs to be effectively reduced in order to improve workers' safety and work efficiency. There are two approaches to noise suppression. The first is passive noise control (PNC), which uses structures that physically block or absorb sound. However, PNC offers limited soundproofing performance, and can be very expensive. The second option is active noise control (ANC), which cancels the unwanted noise from the source sound by generating anti-noise signals which have the opposite phase to the noise. ANC is well-known to be effective at removing noise in the low-frequency band, but its traditional algorithms tend to be unreliable filters that do not accurately capture noise sources that are nonlinear and have complex patterns. To address this issue, filters based on deep neural networks are emerging as an alternative approach, because they are known to enable effective modeling for nonlinear systems. Noise suppression using convolutional neural networks (CNNs) as an acoustic model has been applied mainly in automatic speech recognition and voice quality enhancement. However, the development of deep learning-based noise control for industrial applications has received little attention. In this project, the applicant and Vintec Acoustics aim to develop an ANC method with a new algorithm that uses advanced artificial intelligence techniques (i.e., deep learning) and is specialized in eliminating noise from aircraft cockpits. Deep CNNs are core modules to detect and estimate the spectral features of the noise from an input sound. It will also be necessary to build a new lightweight CNN structure that can reduce computational costs, because aircraft noise reduction requires real-time processing. In an aircraft, any phase shift (time delay) could have catastrophic consequences. It is expected that the system developed in this project will serve as a template for various industrial fields requiring real-time noise reduction technology.**

Vintec Acoustics专注于为各种与声学相关的工程问题提供解决方案，包括高性能音响设备设计和噪声和振动控制。建筑工地、汽车或飞机等工业领域会产生持续的噪音，为了提高工人的安全和工作效率，需要有效地降低这种噪音。噪声抑制有两种方法。第一种是被动噪声控制（PNC），它使用物理阻挡或吸收声音的结构。然而，PNC提供有限的隔音性能，并且可能非常昂贵。第二种选择是主动噪声控制（ANC），它通过产生与噪声相位相反的抗噪声信号来消除源声音中不需要的噪声。众所周知，ANC在去除低频噪声方面非常有效，但其传统算法往往是不可靠的滤波器，不能准确捕获非线性和复杂模式的噪声源。为了解决这个问题，基于深度神经网络的滤波器正在成为一种替代方法，因为它们能够有效地对非线性系统进行建模。以卷积神经网络（cnn）为声学模型的噪声抑制主要应用于语音自动识别和语音质量增强。然而，基于深度学习的噪声控制在工业应用中的发展却很少受到关注。在这个项目中，申请人和Vintec声学的目标是开发一种采用先进人工智能技术（即深度学习）的新算法的ANC方法，专门用于消除飞机驾驶舱的噪音。深度cnn是检测和估计输入声音噪声频谱特征的核心模块。此外，还需要建立一种新的轻型CNN结构，以降低计算成本，因为飞机降噪需要实时处理。在飞机上，任何相移（时间延迟）都可能造成灾难性的后果。预计本项目开发的系统将作为需要实时降噪技术的各种工业领域的模板