Automatically predicting giant panda mating success based on acoustic features

Abstract

As a solitary species, giant pandas do not frequently vocalize. However, they make significantly more vocalizations during the breeding season, implying that vocalizations are essential for coordinating their reproduction and expression of mating preference. Previous studies have also shown that giant panda vocalizations are correlated with mating results and reproduction. This paper makes the first attempt to devise an automatic method for predicting the mating success of giant pandas based on their vocalizations. Using audio recordings of mating giant pandas collected during breeding encounters, we firstly isolated the vocalizations, and normalized their decibels and duration. We then extracted acoustic features from the audio segments and fed the acoustic features into a deep neural network, which classified the mating result into a success or a failure. The proposed deep neural network employs convolution layers followed by bidirectional gated recurrent units to extract features relevant to mating preference. Evaluation experiments on a data set collected during the past nine years obtained promising results, demonstrating the potential of audio-based automatic mating success prediction in assisting giant panda reproduction.

Keywords

• Giant panda
• Mating success prediction
• Speech emotion recognition
• Deep neural network