Automatic detection of vocalizations of the frog Diasporus hylaeformis in audio recordings

A method for the automatic detection of calls of the frog Diasporus hylaeformis (Eleutherodactylidae) in audio recordings is proposed. The method uses the loudness, timber, and pitch of the vocalizations to identify the calls of the most prevalent individual in a recording. The first step consists in calculating the loudness of the signal to recognize the sections where the focal individual's vocalizations are. The second step consists in using the timber of the signal to recognize vocalizations. Finally, we use two principles we observed in the sounds produced by this species to discriminate between the calls of the most prevalent individual and other calls: individuals tend to vocalize using an almost constant pitch and different individuals use different pitches. Results show that the method is resistant to background noise (including calls of individuals of the same species), microphone-manipulation-induced noise, and human voice, and also that it adapts well to variations in the microphone level produced during the recording.