基于FFT变换和ADALINE网络的实时音频滤除窄带噪声方法  

Real⁃time audio narrowband noise filtering method based on FFT and ADALINE network

作　　者：谢家兴 梁高天 尹东晓[4] XIE Jiaxing;LIANG Gaotian;YIN Dongxiao(College of Electronic Engineering,South China Agricultural University,Guangzhou 510642,China;Guangdong Modern Agricultural Science and Technology Innovation Center for Intelligent Orchard,Guangzhou 510642,China;Guangdong Engineering Research Center for Monitoring Agricultural Information,Guangzhou 510642,China;Luoding Polytechnic,Yunfu 527200,China)

机构地区：[1]华南农业大学电子工程学院,广东广州510642 [2]广东省智慧果园科技创新中心,广东广州510642 [3]广东省农情信息监测工程技术研究中心,广东广州510642 [4]罗定职业技术学院,广东云浮527200

出　　处：《现代电子技术》2022年第9期41-45,共5页Modern Electronics Technique

基　　金：广东省科技专项资金(“大专项+任务清单”)项目(2020020103);广东省教育厅特色创新类项目(2019KTSCX013);广东省现代农业产业技术体系创新团队建设专项资金(2020KJ108);广东大学生科技创新培育专项资金项目。

摘　　要：针对传统的噪声滤除方法实时性不好,需要人工设置的参数过多导致使用不便的问题,利用快速傅里叶变换(FFT)获取噪声音频的频谱,选取频谱中幅度最大的前若干个频率作为噪声频率,依据噪声频率重构初相为0的正弦波噪声样本,利用改进自适应ADALINE网络模拟噪声音频中的噪声,将噪声音频减去网络模拟出的噪声从而实现音频去噪,同时把相减结果作为误差反馈给网络用于权值更新。相比传统方法,该算法不仅处理时延更低,而且只有一个参数需要设置,大大简化了使用流程。仿真测试发现该算法可以有效滤除高斯窄带噪声,滤波后信噪比提高了11 dB,并且算法实时性强,初始化仅需82 ms,音频采样点处理速度高达102 kHz。In view of the poor real⁃time performance of the traditional noise filtering methods,which have too many parameters that need to be manually set,and are inconvenient to use,FFT(fast Fourier transform)is used to obtain the frequency spectrum of noise audio,several frequencies with the largest amplitude are selected as the noise frequency,the sine wave noise sample with the initial phase of 0 is reconstructed according to the noise frequency,the improved adaptive ADALINE network is utilized to simulate the noise in the noise audio,and the noise generated by the network simulation is subtracted from the noise audio to achieve audio denoising.The subtraction result is taken as an error to feed back to the network for weight update.In comparison with the traditional methods,this method not only has lower processing delay,but also has only one parameter to be set,which greatly simplifies the application process.It can be seen from the result of the simulation test that the method can effectively filter out Gaussian narrowband noise,and the signal⁃to⁃noise ratio after filtering has increased by 11 dB.The method has strong real⁃time performance,its initialization only takes 82 ms,and the audio sampling point processing speed reaches up to 102 kHz.

关 键 词：窄带噪声滤除 FFT ADALINE网络 噪声音频获取 噪声重构 音频去噪 仿真分析 

分 类 号：TN911.4-34[电子电信—通信与信息系统]