The sound spectrograph (often termed 'sonograph'), analogical or digital, has been the instrument of choice for the analysis of animal vocal signals. This instrument gives effectively a useful time-frequency representation. Nevertheless, because a signal varies simultaneously in two-dimensions, time and frequency, spectrographic measurements are constrained by the 'uncertainty principle': to increase accuracy of measurement in one dimension, it is necessary to sacrifice accuracy of measurement in the other dimension. There are, however, a number of alternative ways to analyse signals, particularly for non-periodic, frequency modulated signals, in which frequency changes rapidly over time, or for signals produced with two independent voices, phenomena relatively common in bird vocalizations. Different methods are zero-crossing measurements, instantaneous frequency measurements by Hilbert transform, pitch detection by autocorrelation or by cepstrum and Wigner-Ville and wavelet transforms. Their respective advantages and limits can be assessed using some examples taken from bird vocalizations studies. These alternative ways of measurement often minimize the uncertainty principle and thus permit a more accurate analysis of the instantaneous frequency of non-stationary signals.

Abstract: