Meeting Abstract
Measuring how similar two sounds are to each other has diverse applications in biology, such as quantifying imitation in vocal learners or reconstructing patterns of sound variation across populations and species. A powerful way to measure sound similarity is to calculate the statistical dependence, or mutual information, between the power spectra of the sounds being compared. Mutual information allows the researcher to measure similarity even when spectra are not linearly correlated. Non-linear dependencies can occur, for example, when two complex sounds differ because one of them displays additional elements or some of its elements are slightly shifted in pitch relative to the other sound. In those cases, using measures that assume linearity may lead to misleadingly low values of similarity. Here, we present an R package that allows measurement of sound similarity via mutual information. The package allows the researcher to automatically compare hundreds of sounds in one batch. As an example, we present measurements of the accuracy with which several juvenile zebra finches (Taeniopygia guttata) imitated the vocalizations of adults. The package also allows the researcher to determine the extent to which elements from multiple sounds are represented in one composite sound. We used this feature to test the extent to which juvenile zebra finches imitated the song of one tutor versus that of a second tutor. This package is expected to increase the use of mutual information to measure similarity in natural sounds by making this method more accessible to biologists.
