Environmental Sound Recognition Enhancement Algorithm based on GAN

Aiming at the problem that the accuracy of environmental sound classification model is affected by the lack of sufficient labeling data, a data enhancement method based on GAN network is proposed. The data set expanded by GAN network data is used to improve the performance of environmental sound classification network. Firstly, the structure of generator and discriminator in GAN network is constructed. Then, the generated audio and the original audio are extracted by Meir spectrum feature. Finally, the extracted Mayer spectrum features are sent into the constructed convolutional neural network for classification and recognition. In the experiment, four traditional data enhancement methods, namely, time stretching, tone shifting, dynamic range compression and adding background noise, are selected to carry out comparative experimental analysis. With UrbanSound8K data set as the experimental object, the proposed model is comprehensively verified. The experimental results show that the proposed enhancement method performs well in the task of environmental sound classification, and the classification accuracy is up to 93.3%.