Classification of Noise Between Floors in a Building Using Pre-Trained Deep Convolutional Neural Networks
Hwiyong Choi, Seungjun Lee, Haesang Yang, Woojae Seong · 2018
This paper suggests a method for source location and type classification of noise between floors at an apartment complex, which is a serious social conflict issue in Korea. Pre-trained convolutional neural networks proposed by visual geometry group is adapted and used for the task. A dataset for evaluation of method is generated and gathered in a building. The dataset is converted to log scaled mel-spectrograms to be fed into the input of the networks. The method is evaluated via k-fold cross validation. For comparison of performance depending on network architecture, convolutional neural networks suggested by Salamon and Bello [IEEE Signal Process. Lett. 24, 279-283 (2017)] is employed and validated. Also, the effectiveness of pre-training is measured.