The Cacophony Project is a mix of technical innovation and conservation. We’re designing our own hardware, collecting a huge amount of data, looking for new ways to analyse sound, and building apps and software to show the world what we discover. If we get it right, we’ll discover the best ways to protect New Zealand birds.
We’re testing the assumption that sound can be a pretty good, high-level indicator of what is going on in an area of bush. At its most basic: Better conservation = more birds = more sound.
The steps are to record sound in the wild, combine it with other data (e.g. GPS coordinates and time), upload it to the cloud (we’re already using Amazon Web Services), process and analyse everything, and produce useful insight into the health of our native bush over time.
Every step comes with questions that you can help answer.
Data collection hardware – Cacophonometer
The aim of this project is to have a cheap reliable way of collecting and uploading data to the cloud. Our initial assumption is that sound is a pretty good high level indicator of what is going on in local area.
A cheap smart phone is the ideal first Cacophonometer and it has all the required electronics. We put the cacophonometer in a water proof container with a plug in microphone and solar charger. This give us an inexpensive way to capture the state of the environment in lots of places. …Detailed specification of initial data logger is here
We are keen on any contributions that will help create a cheaper, more reliable device.
Recording sound in the wild - the Cacophonometer
Put simply, the Cacophonometer is a cheap sound recorder that can survive outdoors, living off solar power and uploading data to the cloud. It’s based on an Android smartphone, sealed in a waterproof container with a plug-in microphone and solar charger. It’s relatively inexpensive and has let us start capturing data in a few different places.
We’ve already built the first version, and now we’re looking for ways to make it cheaper and more reliable.
The smartphone base gives us some obviously useful components - like a core processor, microphone and GPS inputs, and wifi/cell/Bluetooth connectivity. But there are questions we haven’t answered yet. What could we do with the camera, or other sensors?
We picture thousands of Cacophonometers listening to the bush all over New Zealand in the future. Could you help create a cheaper, more reliable device?
You’ll be able to buy Cacophonometers soon. Leave us your email address and we’ll let you know when.
- Core processor – e.g. Arduino or cheap smart phone
- Data input devices – sound recorder, GPS, temp
- Data transfer devices like wifi, cell data, Bluetooth, SD card
- Power supply – solar collector
Buy a Cacophonometer and see what it can do - and what you can do for it!
Uploading and storing data
Cacophonometer software is upgradable, so as we learn we can improve the way we sample, process, and upload data. We might even find that different configurations suit different locations or objectives. We’re looking for data-heads to help us work out:
- Data sampling - How often should Cacophonometers record sound samples? At what quality?
- Data processing - What processing is best before we upload? Should we compress to MP3 to reduce data size?
- Upload methods - We’re using wifi or cell data for now, but we want to look into Android’s open source mesh network software, especially where there’s no phone reception (i.e. most of the back country).
- Data cleaning - What do we need to remove from our recordings? How can we detect and clear things like voices, which we want to exclude for privacy reasons?
- Storage structure - We’re storing this data for everyone, and want it to be useful for decades to come. Will a simple structure be best? Sound recording, timestamp, GPS coordinates - is that all we need?
Our biggest goal is for this data to be available for analysis for as long as possible. Moore’s Law tells us that amazing things will be possible in the future, and if we’re flexible enough now, today’s data can fuel tomorrow’s discoveries. The biggest ‘unknown’ is what will be possible, or how it will happen.
Some analysis we’d suggest for starters:
- ‘Cacophony index’ - The first goal is a simple health index for ecosystems, based on sound analysis. This will never be perfect, but it could be an objective measure that can be retrospectively analysed as techniques improve. It might be that we listen every 5 seconds for bird noise, and assign a score (from 0-100) on the quantity of bird noise in any given period.
- Trends at each location - Measuring changes in the Cacophony index over time at each data collection point.
- Analysis of pest noises - Can you detect pests like possums or rats based on sound?
- Analysis of control methods - Can you detect a trap going off?
- Analysis of bird species - How accurately can you identify species by analysing recorded bird calls?
- Maps - Layering the Cacophony index, or other trends and data, over a map to visually identify areas of high cacophony, and dead zones.
- Experiment design - Can our data and insight help design better experiments for pest eradication? We could, for example, monitor areas before and after trapping or poisoning, and also collect ‘control’ data from a similar area without traps or bait.
- Optimising data collection - From all of the above, and perhaps other analysis, can you estimate the optimal spacing and spread of Cacophonometers? Can you help optimise other data collection methods?
Apps and software
We want the New Zealand public to play a massive role in the Cacophony Project (and in conservation in general). So let’s engage them through those phones and computers that everyone spends too long looking at. Some people will buy and monitor their own Cacophonometers, while others will want to see and understand the data.
Imagine apps that people could use to:
Digital Lures - sound and possibly images recreating sexual or social behaviours
Digital lures have the following advantages:
- lure can be infinitely long life (solar powered) if sound and or video. Given the labour requirement to rebait etc this could make a very large difference to the total cost over years. The lure could be possibly last 100 times longer.
- can act over large distance (much larger than food based). This is particularly important as 10 times distance cover 100 times area - only 1/100 number of lures/traps needed.
- can be adapted to target specific pests (rats, possums, stoats, mice, rabbits, feral cats, goats, pigs, etc.)
- can be updated to get better over time (as methods get improved)
- network of devices could adapt to target one pest if invasion is noticed (eg instead of cycling through different pest lures just focus on stoat if it has been identified).
- could be set up to eradicate one set of pests at a time eg possums first until all gone, then stoats, then rats then mice.
- less likely to attract non target animals and birds (probably actively avoid).
- becomes even more effective in low density situations (unlike food lures that become less attractive in food abundant environments). Finding a mate is harder in low density so may try anything that vaguely looks promising? This may also mean smell is less important as there are fewer competing smells around.
- field devices can listen for response and adapt responds to lure pest in. For example there may be a series of communications back and forward that changes. Also possible that you could have a highly amplified sound to cover larger area and then move volume down as animal responds. This could vastly reduce the number of traps required for a given area.
Ultimate digital trapping system
- Low cost solar powered smart phone is core of the intelligence and communications.
- Sound and or video is the core of the sexual/social lure (ultrasonic or pheromone sprays could be added too).
- Device can listen for any of the possible pest out there and when detected the entire network adapted to target the one pest.
- Pest identification at the trap is done with image recognition (getting vastly cheaper).
- Kill is done by any number of killing methods eg squirting poison (solar powered propulsion), poison dart, gas – the pest can not become trap wary as there is nothing to go into. Plus they go elsewhere to die so would not need to be cleared.
- All open source so anyone can buy devices and try new variations to improve all systems – engage all the intelligent folks in NZ and possibly world to tackle each part of the problem.
- All data is automatically uploaded to open source cloud database and improvements can be rolled out to all systems and improvements are made eg better digital lure, better at listening for pests, better data on how to time and rotate digital lure, better image recognition, better power management, better way to run the trap seasonally and in different weather, etc.
- Moore’s law means this type of system is likely to make huge improvements in effectiveness and cost relatively quickly. Possibly improving cost or effectiveness by a factor of two every few years.
Open Source - Participation and Reuse
Our main goals with this project are to:
- create a community of interested technologists, biologists, and other researchers around the hardware, software and scientific problems we need to address to make the project a success, and
- encourage commercial interests to provide services which dovetail with or benefit from the work we are doing.
We are working to ensure that our licensing first and foremost protects the interests of our community and collaborators. Our next priority is to be an attractive partner to commercial interests.
The software components of the Cacophony Project are made available as "Free Software", which is also open source, under copyleft licenses like the GNU General Public License, v3 or later (GPL) or the Affero General Public , v3 or later (AGPL). We do this to protect the interests of both users and developers, who are often the very same people.
To the extent to which we are able, we will entertain requests to make individual software components available to commercial partners under different licenses which allow commerical exploitation, however we will do this only if a sufficiently strong, mutually beneficial case for doing so can be made, as these sorts of arrangements can strain the relationships of open source communities.
This project is built around a small set of open source collaboration tools: OnlineGroups.Net for email list communication, Git for source code control, and Taiga for Kanban-style project coordination. All of these tools are both open source and gratis (free of cost) to use.
If you want to find out more about the project - regardless of whether you're a software developer, scientist, or just an interested member of the the public, we invite you to join our OnlineGroups.Net mailing list (provided by the NZ Open Source Society) and get involved in the discussion, or just browse/search the past discussion in the archives! It's the easiest way to participate, and participation is what we're after!
The main project, with its software sub-components is the Cacophony Project on GitHub. You are welcome to peruse the existing code base, and we encourage you to test our code - feel free to fork away! We love pull requests.
We manage the day to day work on the project via our Cacophony Project on Taiga.io (Taiga is similar to Trello, but is free to use, and fully open source, i.e. you can create your own self-hosted instances of Taiga) - you currently need a user account (available at no cost) on Taiga to join the project.
We have selected technologies for a number of the planned components of the Cacophony Project, and number are still in the planning stages. Technology choices will be made based on what our committed contributors believe offers the best likelihood of success and future viability. We will, of course, only use free and open source technologies to ensure that all interested contributors are able to participate without any artificial barriers.
For our Cacophonometers, we are using the following technology stack:
- Android (4.0 or better) OS
- Android SDK and developer tools
- Commodity hardware
For server components, including our sound file aggregation and distribution service, we are planning to use (subject to compelling arguments for other technologies):
- Linux Server (virtual instances) - Ubuntu 14.04 LTS
- Amazon S3 for sound data storage