Acoustic surveillance of hazardous situations using nonnegative matrix factorization and hidden Markov model

Abstract

In this paper, an acoustic surveillance method is proposed for accurately detecting hazardous situations under noise conditions. In order to improve detection accuracy, the proposed method first tries to separate each atypical event from the input noisy audio signal. Next, maximum likelihood classification using multiple hidden Markov models (HMMs) is carried out to decide whether or not an atypical event occurs. Performance evaluation shows that the proposed method achieves higher detection accuracy under various signal-to-noise ratio (SNR) conditions than a conventional HMM-based method.