Automated ambient recognition method based on dynamic center mirror local binary pattern: DCMLBP

Voice recognition and sound classification is a hot-topic research area in the literature and many methods have been presented. Ambient recognition is very important problem by using voices or acoustic. Especially, digital forensics examiners need an automated ambient recognition system by using acoustical voices. In this study, a novel automatic ambient recognition method is presented by using acoustic. Firstly, an acoustical voice dataset is acquired. These voices are categorized in the 8 classes and these classes are kinder garden, ferryboat, airport, cafe, subway, bus, traffic and walking. Then, a sequential learning method is presented for ambient recognition using acoustical voices. The proposed method consists of dynamic center mirror local binary pattern (DCMLBP) and discrete wavelet transform (DWT), neighborhood component analysis (NCA) based feature selection and classification phases. By using DWT, a sequential learning method is proposed and the proposed feature extraction method has nine levels. Experiments clearly show that the proposed DCMLBP based method has high classification accuracy, precision, geometric mean, F-score for ambient recognition. According to results, the best accuracy rate was calculated as 99.97% +/- 0.07% by using support vector machine and 128 features. (C) 2019 Elsevier Ltd. All rights reserved.