Bioacoustics Signal Classification Using Hybrid Feature Space with Machine Learning

Abstract

The vocal sounds emitted by animals and birds possess distinctive signatures. They are vital for acoustic monitoring to extract useful ecological data and track biodiversity. Recently, automated bioacoustics classification drew attention from the research community due to its diverse application. To that aim, we present a novel classification method for acoustic data by fusing optimally selected signal features. The proposed method extracts the distinctive statistical features, calculated using coefficients of Discrete Wavelet Transform (DWT), and fuses them with the descriptive features estimated using the Mel-Frequency Cepstral Coefficients (MFCC). The combined feature set is then passed to different machine learning classifiers, including Random Forest (RF), Support Vector Machine (SVM), and k-Nearest Neighbors (kNN) for sound classification of different animals and birds. The evaluation results show that the proposed method improves the classification accuracy and achieved high precision on all classifiers.