Audio event analysis for auditory scene recognition

Audio data contains several sound types and is important source for multimedia applications. In this thesis, we present a system for analysing and detecting 16 distinct audio events namely; alert, clear throat, cough, door slam, drawer, keyboard, keys, knock, laughter, mouse, pageturn, pen drop, phone, printer, speeh and switch that are collected from office live environments. The recognition of 10 different auditory scenes (bus, busy street, office, open airmarket, park, quiet street, restaurant, supermarket, tube and tubestation) is also performed in the study. Moreover, correlations between audio scenes and audio events are investigated. Support Vector Machine (SVM) classifier along with the Mel Frequency Cepstral Coefficient (MFCC) feature are used throgh the analyses. In addition, we propose an adaptive frequency analysis scheme for feature extraction and perform optimizations for feature representation and classifier design.