discussion / Sensors  / 17 August 2021

Feedback welcome

Hi All, I'm looking for some feedback from people in the fields of bioacoustics, if you happen to know anyone with an open mind that would be up for a discussion. 

My background is in the use of IoT for environmental conservation, previous work has been the design and manufacture of IoT devices to transfer telemetry from public use drinking fountains to report on the reduction of co2 and single-use plastics in Australia.

I'm currently studying my master of IoT @ La Trobe in Australia & I have been developing an open source add-on to the RAKWireless wisblock system https://docs.rakwireless.com/Product-Categories/WisBlock/ system (modular IoT device) that I think may be useful to enhance the current state of bioacoustics monitoring. My development on the addon was largely influenced by the current state of acoustic recording devices currently available such as the audioMoth & SongMeter, my wish is to enhance the current technical implementation of that type of device whilst targeting a similar price.

So, my current understanding of audio monitoring / bioacoutics as of now is, feedback welcomed.

Current technologies in the realm of acoustic audio monitoring all rely on set and hope method of operation, you install the device and come back and retrieve the data if any, they currently produce large data files that can be arduous to analyze, and the sampling of data is not tailored to the species under study. There is currently no way to retrieve feedback on the quality or validity of the captured data before off-site analysis. Methods of acoustic attractants are used within the ecology community which has been proven successful for attracting species, these acoustic attractants can be abused as there is no control on the frequency and species they are targeting.

I am curious to know if any of these features would be advantageous to researchers, feedback welcomed:

  • The ability to use learning recall / machine learning to identify specific species on the device.
  • Enable species triggered recording whether it be acoustically, or proximity triggered
  • Make use of acoustic lures to validate acoustic anomalies, attract/deter species
  • Remotely verify the quality of sampled data
  • Remotely reconfigure the device as per the environment.

Thanks.




Hi adamjp.

I'm not a specialist in Bioacoustics and it sounds like that is the feedback you are looking for. The features you're proposing sound like they'd be welcomed. If this will become a product, perhaps I can give some advice from a product design point of view though.  I think in addition to the features, you may want to discuss documentation, support, availability, price, reliability, and usability with people in the field. I also recommend working with domain specialists in developing it so you can get a sense of which features are most important, how difficult it would be to use or set up, etc. In the past, I've wasted a lot of time implementing things that I thought were important but were not really important to people using my designs. 

Akiba 

Hi Adam--

I think there are a few things to think about here. First is that existing acoustic recorders (Song Meters and Audiomoths) are designed in part to be pretty power-efficient and handle long deployments on relatively basic power delivery (D-cell batteries for song meters, AAs for audiomoths). Any addition of capability is going to have to be balanced with power draw which, for researchers doing long-term studies, is a really big deal. 

I haven't done work with it but I think both Song Meters and Audiomoths can do triggered recording but this is mostly restricted to bats and mainly a consideration given that ultrasonic recordings are extremely storage-intensive for a given duration.

Verifying quality of sampled data, or even sensor status, would be really useful and a great first step since you don't need much bandwidth. Even a nightly packet of "Hey, I'm still working, I have 32GB of storage left and estimate 15 days of remaining deployment left in my batteries" would be a huge step forward over what we have now.

Remote reconfiguration is something wildlife acoustics has already implemented on the SM Mini and Micro, through use of a bluetooth connection to a mobile app. I've found it only moderately useful, the nicest part of it being automatically pulling an up to date GPS point and timestamp from the onboard GPS on my phone.

Hope this helps

Hi Adam,

I build audio recorders and I'm going to echo much of what's been said above.  Power and capacity aren't a big deal, recorders draw 30mA or better, so a dozen D cells and a 64G SD card will record for months very deterministically.  As you observed, you end up with a lot of data, but it's a good problem to have.

So running a classifier on the acoustic device is valuable if it is also paired with notification.  I'm working on blast and gunshot detection/notification using LoRa.  A big part of the problem is having to set up the wireless infrastructure.

One area that David mentioned is in recording ultrasound and the amount of data they generate.  One technique that can be employed to advantage is to downshift the signal (the way bat detectors do it) before writing to SD card.  There is little concern that this kind of "lossy compression" can lose information, since it's generally very quiet in those frequencies.  I'm not sure why this isn't done more.

Thanks, -harold

Hi Adam,

I haven't done energy harvesting but I will need to in order to support wireless (higher power draw plus indefinite deployment duration).  Energy harvesting would also be useful for plain recorders where the deployment area is rugged and one would rather not schlep heavy or bulky kit up and down mountains.  For me, energy harvesting means PV panels at the moment, and  I don't see many stumbling blocks in implementing this.

Regarding ultrasound, heterodyning is one way to downshift the frequency, and it would be most suited when you know the frequency band of interest.  Frequency division is also possible.  Yet another possibility is to calculate the spectrum and record that instead of the time series.  Each of these methods discards some information that would otherwise be present in the time series, but I'd argue the important stuff is still there.  Of course it depends on exactly what the research question is, but often it boils down to getting the spectrogramme.  In any case, this requires way different hardware or software than regular recording, and maybe this lack of familiarity is why it hasn't caught on.

Thanks, -harold