Your smartphone could be so much more than a glorified alarm clock and a portal into the world of Angry Birds. It could also help scientists unlock some of the mysteries of the natural world. In years to come it might even contribute towards protecting global biodiversity, through crowdsourcing.
Crowdsourcing has revolutionised many areas of scientific research, providing data across previously unobtainable temporal and geographic scales. It is hoped that crowdsourcing will be able to help gather information on a particularly poorly studied group of animals that few of us ever see or hear – bats.
Bats are fascinating creatures. Almost exclusively nocturnal, and the only mammals capable of true flight, these enigmatic animals are able to ‘see’ their environment with their ears. Using echolocation, they don’t just fumble their way through the hours of darkness, they dodge, weave and somersault to out-manoeuvre and capture even the deftest of nocturnal insects. But being creatures of the night, and utilising frequencies largely well beyond the range of human hearing, they remain notoriously difficult animals to study. Even with hi-tech bat detectors it is still a skilled job to be able to distinguish one species from another.
So imagine an app on your phone that could automatically identify the dark, cryptic shapes flying figure of eight above your head. No need for years of training, hefty specialised recording equipment and expensive proprietary software – just a smartphone, a free app and an ultrasonic microphone to plug into your headphone jack.
This is what we’ve been working on in the Nature Locator team at the University of Bristol. There is already proprietary software that can identify bat calls for you on a computer. But doing it in real-time, in the field and on your phone with open source software is something no one has managed to date.
BatMobile in action
I would love to say that we’d achieved all of these objectives and it was ready to go. But that (alas) was never our target. BatMobile is a prototype-stage smartphone application. What we set out to do was explore just what was possible. We had many questions to answer. Was the equipment on the market (in this case an iPhone 5) up to the task? (it was). And could machine learning techniques used on a computer be employed to quickly whittle down the likely bat suspects on your phone (they could). But these fundamental (and slightly dull) questions aside, and accepting that the project was only exploratory in nature, what can the app actually do?
Well, in summation, it really can identify bats from their calls. But it can really only do it well in what I would call ‘controlled’ conditions e.g. using clear call sequences. For those of you with a desire to know what’s involved in this ’call in one end – ID out the other’ process here’s a quick breakdown:
- The app employs algorithms that analyse audio files of bat calls. These algorithms are able to isolate (from background noise) any segments which have been identified as potentially being of ‘bat origin’. Clever.
- The app then does something even more clever called parameterisation. What this really means is that it is able to characterise each call segment (bat squeak) it finds. It’s the mathematical equivalent of saying the call has, for example, a maximum and minimum frequency of particular values (e.g. 115 kHz and 94 kHz respectively).
- These identified and characterised features are then passed to a machine learning system that has been ‘trained’ to identify the bat species in question based on the values of these features.
It's all clever stuff. Thankfully, I didn’t have to understand half of it. However, currently our system is not quite robust enough to function adequately out in the field. Reliably performing automatic identification based on audio analysis depends on there being sufficient features to base the analysis on, so that you can discriminate between similar calls, or calls where there isn't much data due to poor recording quality or interference. We currently look at six features in any call including minimum frequency and bandwidth. But for the system to be able to cope with the nuances of the real world at least twice this number of features need to be examined, and perhaps many more.
Overall we have been happy with the progress made in this challenging area. We have shown that modern smartphones are up to the task of handling the analysis of the data and we have shown that the system works well in a controlled environment.
The next step is to refine and enhance the process. And in the near future you could have an app in your pocket that will allow you to eavesdrop on and better understand a previously hidden world. Obviously to those with an interest in bats this is fascinating in its own right but perhaps more crucially it represents what could be an important step forward for nature research and conservation. By knowing more about the distribution and population sizes of our bat species we will be far better equipped to safeguard their future.