Computer-aided analysis of acoustic parameters: new possibilities of signal analysis [abstract]

L. Schrader & K. Hammerschmidt (1996). Computer-aided analysis of acoustic parameters: new possibilities of signal analysis [abstract]. Bioacoustics, Volume 6 (4): 307

Nowadays, hardware as well as software for the analysis of acoustic signals is available at reasonable rates. 0ne can choose between several powerful signal analysis systems, embracing calculation of spectrograms (often real-time spectrograms), power spectra, envelopes and arithmetical or statistical tools, e.g. for the calculation of similarity indices between signals. In addition, it is often possible to make measurements directly on the screen by means of a cursor. Algorithms for calculating specific sound structures, like the fundamental frequency or formants, are often integrated in software for speech analysis, but are less common in software predominantly used by bioacousticians. This is due to the following reasons: acoustic signals of animals can differ drastically in sound production (voiced or unvoiced, tonal or atonal, low- or high-pitched) between species as well as within a species; in addition, recordings are often done under conditions resulting in variable signal-to-noise ratios; furthermore, different environmental influences can filter or mask important cues of a signal. Hence tools appropriate for human speech analysis are often not applicable to the analysis of animal vocalizations. Our approach deals with these problems by means of a high program flexibility. Most calculations are done on the basis of digitized frequency time spectra. Dynamic thresholds are used to separate a signal from background noise and to emphasize certain sound structures. To compensate distortions of signals, heuristic assumptions are integrated. For example, the range of values for the fundamental frequency are limited. We will especially introduce methods to estimate the fundamental frequency, formant structures, dominant frequency bands, and the statistical distribution of spectral energy. 0n the basis of these structures, parameters can be calculated which describe the structure of vocalizations numerically. As a result of this approach a variety of signal structures can be described by a large number of parameters. This has several advantages: first, different parameters might be useful to characterize different sounds, as signals can differ in crucial features, e.g. tonality; second, parameters that cannot be estimated (e.g. fundamental frequency) may correlate with other parameters (e.g. distribution of spectral energy) and can thus be substituted in this way; third, it is often not possible to know the decisive parameters in advance. The analytical methods we propose were tested for vocalizations of different species (including birds, monkeys and pigs) and were found to be appropriate for investigating questions on covariations of acoustical features with individuality, social context, referentiality and internal state.