Acoustic identification of insects based on cepstral data fusion and hidden Markov models

Abstract

© 2021 Elsevier Inc.This study presents an intelligent system for automatic acoustic classification and verification of insect sounds based on hidden Markov models. We propose a new approach to acquire knowledge of this biodiversity through a digital signal-processing technique designed specifically for the identification and verification of acoustic sounds of insects using mel frequency cepstral coefficients and linear frequency cepstral coefficients. The two types of coefficients are evaluated individually, as shown in previous work, and a data fusion is proposed, showing an outstanding increase in identification and classification accuracy, reaching 98.38% on the level of specific species, out of 250 species reported in the singing insects of North America collection and 26 species of katydids from Costa Rica.