Real-time Identification of Respiratory Movements through a Microphone

Abstract

This work presents a software application to identify, in real time, the respiratory movements -inspiration and expiration- through a microphone. The application, which has been developed in Matlab and named ASBSLAB for the GUI version and ASBSLABCONSOLE for the command-line version, is the result of a research and experimentation process. A total of 48 minutes of breathing movements from four subjects was recorded and 18 acoustic features were extracted to generate the data model. A first level of identification, based on the classification of tiny audio segments, was designed using kNN supervised method. The second level of identification implements a state machine that takes the results ordered in the time from kNN as input and identifies the whole respiratory movement, achieving a level of positive identifications above 95%. As computation time is a handicap, the application let the user choose easily the sample rate, the audio segment size and the set of acoustic features to use in the identification process. In addition, based on the number of features selected, this works suggests those that achieve best results.