ERANNs: Efficient residual audio neural networks for audio pattern recognition

Audio pattern recognition (APR) is an important research topic and can be applied to several fields related our lives. Therefore, accurate efficient APR systems need developed as they are useful in real applications. In this paper, we propose a new convolutional neural network (CNN) architecture method for improving the inference speed of CNN-based tasks. Moreover, using proposed method, improve performance systems, confirmed experiments conducted on four audio datasets. addition, investigate impact data augmentation techniques transfer learning systems. Our best system achieves mean average precision (mAP) 0.450 AudioSet dataset. Although value less than that state-of-the-art system, 7.1x faster 9.7x smaller. On ESC-50, UrbanSound8K, RAVDESS datasets, obtain results with accuracies 0.961, 0.908, 0.748, respectively. ESC-50 dataset 1.7x 2.3x smaller previous system. For dataset, 3.3x We name "Efficient Residual Neural Networks".