Dynamic programming matching as a simulation of budgerigar contact-call discrimination

Dynamic programming (DP) matching has been successfully used to classify budgerigar contact calls into natural groups. In this paper, the biological relevance of the DP-matching algorithm was further investigated with an enlarged test set and using psychophysical test results from actual birds. Five call comparison methods (CCMs), including an improved DP-matching method, were used to quantify the dissimilarity or similarity between the contact calls of different budgerigars. Peak-time matrices, consisting of the three most intense peak frequencies in the respective time bins of given calls, formed the common imput to the CCMs. The CCMs were compared in terms of (1) efficiency for classification (cluster analysis) of very similar cagemate contact calls into natural groups, i.e., by individual, and (2) correlation to the birds' own perception as inferred from psychophysical tests. The DP-matching method, which chose two representative peak frequencies out of the three for the best match, proved to be the most effective in the classification of cagemate calls and the most correlated with bird perception. Its performance was comparable to that of individual birds. The use of two prominent peak frequencies, time warping, and sequential matching in contrast to correlational computation were found to be important factors not only in classifying calls into natural groups, but also in simulating the perceptual discrimination of budgerigars. Possible counterparts in the bird's brain, such as those responsible for sequential processing, which could serve as the mechanisms of call recognition, were discussed in connection with the above results.