Automatic Environment Sounds Classification Using Optimum Allocation Sampling

Abstract

Sound provides highly informative data about the environment. In the sound recognition process, the signal parameterization is an important aspect. In the present work, a new approach using optimum allocation sampling (OAS) method based features used in multi-class least square support vector machine classifier (MC-LS-SVM) is proposed for environmental sound classification (ESC). The time and frequency (TF) features are extracted from the OAS method and these features used as input to MC-LS-SVM classifiers with different kernel functions for automatic ESC. Various performance parameters are computed with Cohen's kappa value being 0.8381 and sensitivity, specificity, F1-score, error and Matthew correlation coefficient are 85.42%, 98.38%, 0.854, 14.57%, 83.81% respectively. The adaptability and accuracy of the proposed is better as compared to the previously existing methods on the same data-set. © 2019 IEEE.