Audio-based Event Recognition System for Smart Homes

摘要

Building an acoustic-based event recognition systemudfor smart homes is a challenging task due to the lack ofudhigh-level structures in environmental sounds. In particular,udthe selection of effective features is still an open problem.udWe make an important step toward this goal by showing thatudthe combination of Mel-Frequency Cepstral Coefficients, Zero-udCrossing Rate, and Discrete Wavelet Transform features canudachieve an F1 score of 96.5% and a recognition accuracy ofud97.8% with a gradient boosting classifier for ambient soundsudrecorded in a kitchen environment.