Acoustical respiratory signal analysis and phase detection

In this paper we propose a statistical modeling approach for phase detection of normal breathing sounds. Previous studies have been considering only the detection of inspiration mid-points and breathing onset. Here we focus on the detection of both inspiration and expiration phases. Based on an accurate statistical study of breathing signals, we suggest a nomenclature of respiratory cycle in a modeling perspective by adding a transitional phase between the inspiration and expiration phases. Thus, we put forward a new processing chain using improved Markov model in a bayesian framework in order to segment the signal and to detect the phases. We adapt the recent triplet Markov chain by exploiting priors on the respiratory cycle structure. Experiments on real respiratory signals show encouraging results.