Audio-based Event Recognition System for Smart Homes

Title: Audio-based Event Recognition System for Smart Homes

Authors: Anastasios Vafeiadis, Konstantinos Votis, Dimitrios Giakoumis, Dimitrios Tzovaras, Liming Chen and Raouf Hamzaoui

Abstract: Building an acoustic-based event recognition system for smart homes is a challenging task due to the lack of high-level structures in environmental sounds. In particular, the selection of effective features is still an open problem. We make an important step toward this goal by showing that the combination of Mel-Frequency Cepstral Coefficients, ZeroCrossing Rate, and Discrete Wavelet Transform features can achieve an F1 score of 96.5% and a recognition accuracy of 97.8% with a gradient boosting classifier for ambient sounds recorded in a kitchen environment.

The accepted paper is for the IEEE UIC (Ubiquitous Intelligence Computing) Conference in San Francisco (4-8 August 2017).