Automatic discrimination of beaked whale clicks in noisy acoustic time series

A matched filter bank methodology is investigated for the automatic detection and classification of stereotypical beaked whale chirps in long acoustic time series. Energy normalization and non-dimensional thresholding techniques are employed. A primary confidence level is provided for each detection based on an indirect measure of distortion or malformation of the detection relative to its template waveform. The matched filter correlator peak output value is used to derive a secondary confidence level. Ambiguity experienced by matched filtering is recognized, and removed for some marine mammal clicks by simple time domain or frequency rules. Matched filters are very good at signal detection, but very poor at signal discrimination in some circumstances.