Acoustic ecology is a rapidly growing area that seeks to understand an environment, and the dynamics of animal populations living within it, through ambient sound and the sounds that individual animals make. A typical ecoacoustic survey involves deploying a small number of battery powered recorders that make continuous sound recordings over a period of weeks to months. These recordings are then analysed to identify processes operating in the environment (both natural and anthropogenic), the animal species present, and their behaviour. To date, these surveys have been limited in scale, due to both the cost of the recording devices (often up £1000 each), and also, the manual analysis of the resulting recordings.

The high cost of commercial acoustic recorders is not particularly surprising given that these are custom devices serving a relatively small market. However, in many areas of science, particularly in laboratory work, high-end one-size-fits-all commercial devices are rapidly being superseded by locally developed solutions tailored to address specific settings. This is driven by the emergence of low-cost development boards, such as Arduino and Raspberry Pi, the online maker community sharing designs [indeed, as we're seeing in the WILDLABS.NET community], and online manufacturers offering printed circuit board (PCB) assembly, 3D printing and laser cutting services, which have made low volume manufacturing of custom devices more accessible than ever before.

In the SoundTrap project at the University of Oxford we are exploiting these trends to develop an open-source low-cost acoustic recorder for environmental monitoring applications. Our prototype logger combines a capable low-power 32-bit microcontroller with a sensitive MEMS microphone from a smartphone and an SD-card to store audio recordings. The total cost is less than £20 and the device is capable of recording 18 hours of high quality audio, sampled at 48kHz, using just three AAA batteries for power. To reduce the time consuming task of analysing the resulting recordings, and to maximise the battery life of the device itself, we use acoustic recognition algorithms on the device to decide in real time whether a particular sound is interesting and should be recorded.

To evaluate this prototype, we have deployed it to search for the New Forest cicada (Cicadetta montana); the UK’s only native cicada species. The New Forest cicada has been known in the UK since 1812 but has not been officially sighted for over 20 years. Unlike more common species of cicada encountered in southern Europe whose song is very familiar, the New Forest cicada sings continuously for up to thirty seconds at a frequency around 15kHz. This is close to the limit of most adults’ hearing range, although children can typically hear it clearly from up to 100 meters away. The reason for its disappearance from a small number of well studied breeding sites within the New Forest in unknown; possibly due to climate or land use changes. However, there is still a good chance that colonies still survive in as yet undiscovered sites within the forest, and we hope to discover these.

For the previous three years, in conjunction with the New Forest National Park Authority, the Forestry Commission and BugLife (the Invertebrate Conservation Trust), we have been searching for the New Forest cicada using a citizen science smartphone app which is capable of recognising and recording the song of the cicada. However, after 6000 survey reports from the forest there is still no sign of the cicada. This might be because the cicada is very particular about when it sings, favouring warm sunny days with little wind between May and June, and our citizen scientists just haven’t been there when they are singing. We hope that deploying fixed recorders at the most promising sites, particularly less accessible, sunny, south facing clearings in the forest, will be more successful.

The recorders deployed in the New Forest are able to automatically detect the main frequency components of the cicada’s song. By waking up every five seconds, and listening for 200ms, they can operate for over eight weeks, continuously monitoring an area and making recordings of any cicadas heard. The resulting recordings are collected at intervals and uploaded to a website where we apply more sophisticated analysis. So far during June 2016 we have collected and analysed over 10,000 recordings from the forest, but have yet to see any sign of the cicada.

Following on from the cicada deployment, we are working with a number of other groups to understand the use cases and resulting operational requirements of future devices. For example, we are working with Prof. Kate Jones at University College London to extend the recorders to ultrasonic frequencies for bat monitoring, and with Prof. Patrick Doncaster and Dr. Jake Snaddon at the University of Southampton to develop monitoring tools for areas of protected tropical forests in Belize. These areas serve as corridors for native Jaguar populations but are threatened by the illegal logging of timber and the the illegal hunting of wild boar; both of which disrupt the Jaguar’s environment and its food chain. To this end, we are developing acoustic recognition algorithms that can reliably detect the sound of chainsaws and gun shots in the forest. While chainsaws are somewhat similar to the cicada detection problem in that they are long duration sounds which offer many opportunities for detection, gun shots present a more challenging problem requiring ultra-low power devices that can continuously listen and analyse sound recordings, looking for the very short signature of a gunshot. In addition, while a recorder is a useful first tool for the rangers currently working in the area, a more developed solution will also need a low-power long-range radio capable of providing alerts to the rangers in real time.

Our aim in all these projects is to develop an open platform consisting of both hardware and software components, that can be rapidly and locally re-purposed to any new application, providing a cheaper and more capable solution than current commercial devices.

About the Author

Alex Rogers is a Professor of Computer Science at the University of Oxford. He is interested in developing and applying principled artificial intelligence, machine learning and agent-based approaches within physical sensor systems to address real-world problems focusing on sustainability. His recent work has addressed future energy systems, such as the smart grid, citizen science platforms, and environmental monitoring, and typically involves the real-world deployment of novel approaches in devices, smartphones or the cloud.

Alex is the co-founder of a home heating advice spin-out, called Joulo, that combines a unique low-cost temperature logger with cloud-based analytics. He maintains the GridCarbon app for monitoring the carbon intensity of the UK electricity grid. He am searching for the New Forest cicada using smartphones and is developing a low-cost, open-source, acoustic logger for environmental and biodiversity monitoring (soundtrap.io).