Classification of Snoring Sound-Related Signals Based on MLP

摘要

An efficient method to classify snore, breath sound and other noises based on the multilayer perceptron (MLP) was proposed in this paper. The spectral-related feature sets of the sound were extracted and used as the input feature of MLP. The minbatch training was designed to get the effective MLP model in training process. The dropout method was applied to optimize the structure of MLP. The correct rates for distinguishing snoring, breathing sounds, and other noises are 98.88%, 97.36%, and 95.15%, respectively.