Animal Species Identification Using TESPAR S-Matrices based Features

Identifying animal species based on the sounds emitted during daytime or nighttime activities is of great help in assessing biodiversity. Mainly, other authors used cepstral coefficients to identify animal species. In this paper we evaluate the use of TESPAR analysis applied to a different kind of cepstral coefficients (Teager-MFCC) and additional features (time and spectral features), to generate input data for a classification system. Experiments on 10 species (domestic and wild animals as well as human voice recordings) using this approach shows that TESPAR S-matrices of these features can be used to achieve effective discrimination between species, even with small number of training data (sound recordings) having small lengths and different sampling frequencies.