Audio noise detection using hidden Markov model

Simple noise level monitoring systems, which are currently used to create noise map in residential areas, are unable to identify source of environmental noise. The proposed automatic noise recognition (ANR) system can be used in conjunction with simple noise level monitoring to create an intelligent noise monitoring system (INMS). The presented system which is focused on aircraft noise detection, consists of two parts: feature extractor and training-recognition. We append linear prediction coefficients to Cepstrum coefficients to make a rich feature extractor. The hidden Markov model (HMM) is used for training and recognition. The required observation sequence is obtained by means of a vector quantization method based on fuzzy C-mean clustering. 15 signals are used for training and 28 signals are used in test phase. An overall 83% accuracy in classification is achieved.