Listening to Elephants

I work at a company which employs biologists and ecologists, and I learned recently that they do bird surveys – counting of the numbers and different types of birds by listening to their chirps and birdsongs. I thought they would do that by going out and looking for them, but of course they’re not so easy to see and they’re much easier to hear! Maybe someday I’ll ask about what they listen for and how they identify the birds.

But along those lines, I was a little surprised to find that researchers are counting and identifying elephants using their acoustical signals as well.

For example, other than counting dung piles along a transect to try to estimate population density, the only typical method of collecting data on forest elephants is by direct observation at clearings in the forest, measuring things like sex ratio, visitation frequency and reproduction. Although some of these measurements can’t be made just from acoustic records — at least not yet — relative numbers of elephants and reproductive activity can be measured, and not just at clearings but anywhere.

The Elephant Listening Project at Cornell University seems to be a great project, focusing on the measurement of low frequency noise from elephants, and using the data for various conservation efforts. There’s a good number of visuals and audio playback of their measured elephant noises, as well as some great pages which explain low frequency noise or infranoise in laymans’ terms. I’d suggest listening to these audio clips using headphones, which are generally better than most cheaper sets of computer speakers.

Of particular interest to me are the sensors they are using to measure and record the sounds. Unfortunately, there’s no link to the acoustic arrays they mention in the text.

In spite of the effort needed to carry equipment and power supply tens of kilometers through the forest, the hard part really comes after the forest sounds have been recorded. Acoustic files are ‘real time’, so the challenge is to locate sounds of interest within the matrix of ‘background’ sounds. Developing automated detectors that can recognize and mark the sounds of interest is a major focus of the programming group of the Bioacoustics Research Program. Elephant calls provide a particular challenge because of their very low frequencies.

The development of automated detectors to mark elephant noise specifically in a recording is not a simple challenge. Just look at the visualization on the Sounds of Forest Elephants. How would you get a computer to identify an elephant sound?