Snoring sound classification using multiclass classifier under actual environments
Keisuke Nishijima, Ken’ichi Furuya · 2019
The problem with conventional snoring sound identification methods is that their performance declines when the snoring sound is identified in the actual environment. Therefore, it is necessary to cope with the stationary and nonstationary environmental sounds that cause the decrease. In this research, we tried to cope with stationary environmental sounds by spectrum subtraction method for noise suppression. Non-stationary environmental sounds were regarded as one class for each type of environmental sound. We tried to identify the snoring sounds by multikernel learning, which is a multiclass extension of a support vector machine and by multilayer perceptron, which is a kind of neural network.