基于Conv-TasNet的变压器音频降噪识别网络  

Transformer Audio Noise Reduction Recognition Network Based on Conv-TasNet

作　　者：胡赵宇 李喆[1] 蒙国勇 冯彦维 陈海威 陆忻 Hu Zhaoyu;Li Zhe;Meng Guoyong;Feng Yanwei;Chen Haiwei;Lu Xin(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Guangxi Electric Power Design Institute Co.,Ltd.,China Energy Engineering Group,Nanning Guangxi 530007,China)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]中国能源建设集团广西电力设计研究院有限公司,广西南宁530007

出　　处：《电气自动化》2024年第6期82-85,共4页Electrical Automation

摘　　要：为降低环境噪声对变压器声纹识别的影响,提出了基于卷积时域音频分离网络的变压器音频降噪识别网络。首先使用卷积时域音频分离网络去除环境噪声,然后使用卷积神经网络实现声纹识别。通过故障模拟试验得到变压器音频数据集,并与其他降噪方法对比降噪效果。试验结果表明,所提方法将数据集音频尺度不变的信噪比提高了9.84 dB,识别准确率提高了25.85%,均优于其他降噪方法。在现场应用中,提出的降噪识别网络将误报率降低至1.2%,并成功实现了变压器故障检测。In order to reduce the impact of environmental noise on transformer voiceprint recognition,a transformer audio denoising recognition network based on convolutional time-domain audio separation network was proposed.Firstly,a convolutional time-domain audio separation network was used to remove environmental noise,and then a convolutional neural network was applied to achieve voiceprint recognition.A transformer audio dataset was obtained through fault simulation experiments and the denoising effect was then compared with other denoising methods.The experimental results show that the proposed method improves the scale invariant signal-to-noise ratio of the dataset audio by 9.84 dB and updates the recognition accuracy by 25.85%,both of which are superior to other denoising methods.In on-site application,the proposed denoising recognition network reduced the false alarm rate to 1.2%and successfully achieved transformer fault detection.

关 键 词：变压器检测 声纹识别 声学降噪 声源分离 卷积神经网络 

分 类 号：TM41[电气工程—电器]