Single versus Multi-Source Discrimination in Birdcalls using Zero-Frequency Filtering

In the processing of bioacoustic recordings such as birdcalls, sometimes it is desirable to determine if a recording has one bird calling or has more than one. In this paper, we utilize the well-established zero-frequency filtering method, used for determining significant instants of excitation (also called epochs), for this task. By determining the average number of epochs per second, we are able to reliably discriminate birdcalls made by a single bird from those made by multiple birds. Experimental evaluation on three bioacoustic datasets confirms the reliability of the method. Species identification studies using deep neural network classifiers highlight the utility of the method.