Reconhecimento automático de aves de nomes onomatopéicos utilizando árvore de decisão

Abstract

This work proposes a methodology for automatically determining characteristics, classification and recognition of birds from onomatopoeic names based on analysis of spectrograms of vocalizations of these birds. The filtering and segmentation of the signals are made automatically. Segmentation, which is the separation of the bird vocalization in regions of sound and silence, is performed with the centroid and energy characteristics of the signals. The filtering is made using filters generated using wavelet transform and Chebyshev filters. The recognition of birds is made from a decision tree, constructed automatically based on the following features: entropy, ZCR, centroid, energy, spectral flux, rollof frequency, minimum frequency, maximum frequency and the frequency of greater intensity of the spectrum.