The computer-aided analysis of acoustic signals of mammals is still a problem, as often (a) sound structures are complex, (b) vocal repertoires often comprise an enormous variety of vocalisations, (c) recordings are influenced by the acoustic conditions of the environment, and (d) the distance and spatial orientation of the sender to the microphone changes. In recent software packages for the analysis of acoustic signals, procedures are integrated which allow the calculation of a variety of signal features. However, these algorithms are often problematic under the conditions mentioned above. In this paper, we present a multi-parametric approach which reduces these problems and which allows a quantitative and reproducible analysis of complex animal vocalisations. Our approach comprises the following aspects: (1) reduction of influences of recording conditions, (2) determination of different sound features and (3) calculation of parameters to characterize these sound features. All calculations are done on the basis of the digitized spectrograms. Special attention is given to the use of smoothing algorithms and dynamic thresholds in order to estimate sound features and to reduce influences resulting from recording conditions. The suitability of our approach has been demonstrated successfully for vocalisations of different species.
computer sound analysis, multiparametric approach, vocalisations, mammals, primates