USING VOICE RECOGNITION AS A TOOL IN POPULATION BIOLOGY AND MANAGEMENT

¹ Institut de Recherche sur les Grands Mammiferes, I.N.R.A., Castanet-Tolosan, France
² Equipe de biologie Quantitive, Universite Paul Sabatier, Toulouse

Individual or populational identification of wild ranging animals based on morphological or vocal cues are documented in many avian and mammalian species. Here, we describe a method of automatic voice analysis that may work at an individual level (in this poster a study on Fallow deer) or at a populational level (the example of the Chaffinch).

The method consists of four steps:

1. Recording vocalizations that can be assigned to known individuals or populations.
2. Digitizing vocalizations with an 8 bits analog/digital convertor.
3. Performing an FFT analysis in order to transform each sound file into a power spectrum of 32 variables. This set of variables represents the relative frequency distribution of the sound power.
4. Performing hierarchical clustering, discriminant analysis or neural networks classification on these spectral variables.

In both studies, such analysis provided successful populational or individual discrimination. Classifications were also performed on additional vocalizations of known origin, in order to test the method's reliability. In the case of the chaffinch, the high rate of correct recognition obtained allowed us to use this method for the prediction of the population origin of unknown individuals. Implications of this technique in individual monitoring and/or populational management are discussed.