SOUND RECOGNITION FOR THE DETECTION OF DISTRESS SITUATIONS USING ACOUSTIC PATTERNS

Abstract

Recognizing distress situations is an essential area of research in health monitoring for elderly or frail individuals. When they are alone at home, the likelihood of a harmful fall is elevated. Throughout an accidental fall, the person remains on the floor without receiving any instant aid therefore well-timed reporting to the concerned caregivers is critical. The research community planned several fall detection methods, but certain limitations are still associated with these methods, i.e., computational complexity and high false alarm rate. To resolve these problems, we proposed a study on daily life’s sound recognition to detect fall events at home. In this work, we used a benchmark RWCP dataset with another class of human screaming sounds (male and female) covering 100 sound instances.  Primarily it is a novel approach based on 1-D Local Ternary Patterns (Acoustic-LTP) as feature extractors, along with conventional audio features such as MFCCs and LPCs. Three different types of features are combined to ensure efficient and robust extraction of features for characterizing the acoustics properties of the sound signals. The experimental findings indicate the efficacy of the proposed approach, as classification via SVM attained an accuracy of 98.1%, exceeding results reported in prior studies.