Sounding out diversity with automated identification

Eleutherodactylus juanariveroi
Somebody told me that they visited a museum in Madrid which has a room full of hard drives of recordings. There is one person who spends every day listening to recording after recording. Experts should be spending their time creating these models and working with the results while computers go through the roomfuls of data.
Professor Mitchell Aide
New cyberinfrastructure is allowing scientists to collect, store and analyse acoustic data from tropical ecosystems in order to establish the presence – or absence – of bird, amphibian, insect and monkey species…

Hundreds of thousands of recordings of birds, frogs, insects and monkeys in Puerto Rico and Costa Rica have already been captured using technology developed by the Automated Remote Biodiversity Monitoring Network (ARBIMON).

The system attempts to address one of the major challenges presented by monitoring diversity in tropical ecosystems – that there is often a huge variety of species, some of which are very rare. The technology is described in a paper in open access journal PeerJ this week. spoke to Professor Mitchell Aide, Principal Investigator in the Tropical Community Ecology Lab at the University of Puerto Rico to get the inside track on how this cyberinfrastructure functions and why it is a departure from the norm.

To automatically identify a species in the recordings taken, experts must first create a model using the dedicated web application. They provide examples and ‘train’ the computer to recognise each species from their vocalisations. Once a reliable model has been developed, 100,000 recordings a minute can be examined by the computer for the presence or absence of that species.

Collection can be automated

Of course it will take a while to develop models for all of the species of interest, and there is the disadvantage that less vocal species will necessarily be excluded.

"Any sampling methodology is going to have its limitations, and one of the limitations of using bioacoustics is that you won’t be able to record a species that is not vocal," Professor Aide noted.

"To me, the advantage of this system is that data collection can be automated. At the moment we record for 60 seconds every 10 minutes, which produces 144 recordings per day. That’s nice, but the big advantage is that you can record from permanent stations or leave a machine in a temporary location 24 hours a day for weeks on end; there’s no way you could ever get biologists to do that."

Not only is this method of data collection cheaper and easier, but it makes it possible to record in a number of sites simultaneously. An expert birdwatcher, of course, can only be in one place at one time.

"We’re not trying to eliminate the biologist; we’re trying to get better data for biologists to work with," Professor Aide stressed. "This way, they can sit in their office and analyse recordings from 10 different places.

"Somebody told me that they visited a museum in Madrid which has a room full of hard drives of recordings. There is one person who spends every day listening to recording after recording. Experts should be spending their time creating these models and working with the results while computers go through the roomfuls of data."

Saving time

Although automated recording systems are not a new phenomenon, the combination of software and in situ hardware developed by Professor Aide and colleagues, including Dr Carlos Corrada-Bravo from the University of Puerto Rico’s Department of Computer Science, is innovative in its ability to collect, process and deliver audio recordings from remote locations in real-time to researchers anywhere in the world.

"Our software is different in that it allows people to better manage their recordings and take advantage of the data they have," said Professor Aide. "I had doctoral students before we created this software who had collected 30,000 or so recordings but were only able to use two or three thousand in their final theses because it took so long to process them ‘by hand’."

At the moment the University of Puerto Rico servers are handling the volume of data produced, with both the system and the data stored in the cloud.

"We’ve had to expand as the project grows, with people starting to take advantage of the system and upload their recordings," noted Professor Aide. "One of the things we tried early on was moving everything to the cloud. The cost is disadvantageous, but on the upside there will be no space limitations in the long term."

A permanent, verifiable record

Until such time as models can be created for all known species in a location, the technology will primarily be used for monitoring populations of species which are already known to inhabit an area, Professor Aide explained.

"At the moment we don’t really have any way of pulling out an unknown species from a recording," he noted. "We did take part in a project a few years ago in Hawaii where a group thought there might be a population of a species that had been recorded as going extinct. We put a recording station out in that area and made a model for a closely related species."

The team did not find the extinct species upon examination of these recordings, but this example does highlight the potential of the system to be put to alternative uses. It could be useful, once a new arrival to an area has been detected, to go back to older recordings and pinpoint the moment at which the exotic species arrived, for instance.

"The recordings make up a permanent, verifiable record," Professor Aide pointed out. "If I claim there is some odd species in a place and nobody believes me, they can go back to the original recordings and analyse them for themselves; it wasn’t just an isolated observation that I made."

The ARBIMON team have conducted most of the analyses themselves up to now, testing the system with a number of groups including insects, birds, amphibians and monkeys to demonstrate its usefulness with different species. Their hope now is that many more people will take advantage of the technology and archives in order to study a multitude of species from numerous angles.

"We now have more than a million recordings which anyone can look at and listen to," finished Professor Aide. "The idea is that they’ll be permanently stored there, like museum specimens. People will be able to go back to them 20 years from now and examine them with new analytical tools and ideas."

Read the full text of the paper, Real-time bioacoustics monitoring and automated species identification…



The problem seems to be, James, that one side WANTS religion to be true and the other side DOES NOT WANT it to be true. A genuine search for truth with discussion or debate in a friendly and honest spirit may not solve all the difficulties, but would stop the brawl. This especially applies to my side as we are supposed to love our neighbours.

Commented Graham on
who started it']);">Religion versus science: I don’t care who started it Ltd, Ebenezer House, Ryecroft, Newcastle-under-Lyme, Staffordshire ST5 2UB
Tel: +44 (0)1782 741785, Fax: +44 (0)1782 631856,
Registered in England and Wales  Co. Reg No. 4521155   Vat Reg No. 902 1814 62