In human speech, temporal features have only secondary importance but it seem likely they play a much more important part in avian communication. The different physiology of birds results in many short discrete sounds separated by clear silences. We have used this structure of bird vocalisations as the basis of a number of small linked computer programs. First, and acting as the foundation of all the others, is a program to recognise sound from silence, using amplitude threshold and two glitch filters all easily variable. The times of onset of each sound and silence are stored in a second file that can either be opened in a spreadsheet or accessed for further processing. The initial use of this program was to measure the pulse rates in the fast and slow parts of nightjar churring to allow recognition of individual birds. This program was then expanded to draw 'temporal spectra' of songs (i.e. the relative frequencies of notes of different lengths) and used to explore the temporal differences between the songs of closely related species. The latest development is more ambitious. Each 'note' in a simple song, identified by the first program, is characterised by a few simple time and frequency variables, e.g. note length, frequency mean, slope, curve and complexity. The goal, which is still a long way off, would be to look for sequences using the same type of analysis tools as those developed for word and grammar recognition. However, initial tests with chaffinch song look promising.
Abstract: