Comparison of methods for automated recognition of avian nocturnal flight calls

Abstract

Many species of birds in the Americas vocalize during nocturnal migration flights. Acoustic detection and classification of the calls shows potential for study of the natural history of these migrant birds. In particular, information about the species composition and number of birds involved in migration movements may be obtainable through acoustic techniques. Other methods such as radar monitoring may have capability only to assess the number, but not composition. Here the feasibility of classifying nocturnal flight calls of birds in the family Parulidae using spectragram correlation and using Gaussian mixture models of Mel frequency cepstral coefficient distributions are evaluated and compared. Tests performed on a set of recorded calls show that the techniques are complementary and may, with improvement, enable automated detection.