The automatic classification of environmental noise sources from their
acoustic signatures recorded at the microphone of a noise monitoring
system (NMS) is an active subject of research nowadays. This paper sho
ws how hidden Markov models (HMMs) can be used to build an environment
al noise recognition system based on a time-frequency analysis of the
noise signal. The theory of HMMs is briefly reviewed in the context of
automatic noise recognition. The performance of the HMM-based approac
h is evaluated experimentally for the classification of five types of
noise events (cal, truck, moped, aircraft, train). With more than 95%
of correct classifications, the HMM-based approach is found to outperf
orm previously proposed classifiers which were based on the average sp
ectrum of noise events. A classification rest performed with human lis
teners for the same data shows that the best HMM-based classifier also
outperforms the ''average'' human listener who achieves only 91.8% of
correct classifications for the same task. (C) 1998 Elsevier Science
Ltd. All rights reserved.