Bat Observations including Sound and Spectrograms

I have noticed that bat observations submitted to iNaturalist vary quite a bit. I have done some research and put together my approach to submitting them that I am hoping increases the value to users of this data. I record sound on a pixel 6 smartphone using a Wildlife Acoustics Echo Meter Touch 2. I transfer my wav files to a PC and use the free Wildlife Acoustics software: Kaleidoscope lite to produce a number of spectrograms. I have found that I need to take a photo of the alpha data analysis chart produced but other than that use Kaleidoscope’s screen saver. From there I load into iNat.
I have just started to do this for existing bat recordings but I am wondering if this is sufficient. I realize that the IDs produced have a high error rate but hoping that what I am submitting will allow someone with more experience to confirm and correct what the sensor / software indicate.

Adding spectrograms as images has been discouraged by staff:
https://forum.inaturalist.org/t/spectrograph-allowed/31621/2

But see this feature request:
https://forum.inaturalist.org/t/automatically-add-a-spectrogram-view-to-observations-with-sounds/3579

3 Likes

Yes, please don’t add spectrograms as pictures as mentioned above. I’m including the staff response from one of @thomaseverest 's links here since I think it sums up some of the issues well:

There appears to be some inconsistent guidance on iNaturalist about submitting bat acoustic data in the form of spectrograms and spectrogram metadata. I understand that originally iNaturalist was primarily focused on visual data in the form of photographs but it also from the start allowed sound to be submitted. Spectrograms are just visual representations of sound as many well know so why not sort out how to deal with spectrograms unless there is a consensus that they are not important. I believe they are. Bat morphology is very difficult to document and with most active behavior being at night it is very problematic for most citizen scientists to photograph them. Augmenting observational data using all formats is useful in my view. After all bats are a significant portion of all mammal spices so we should sort out the challenges of recording them.
I have made my own humble attempt to sort out what would constitute adequate acoustic attributes to document in spectrograms and meta data and have started to submit views in my observations accordingly. My workflow is based on a commonly available bat sensor and analysis software. Yes I realize full well that as with all software (including iNaturalist’s AI) there are biases and margins of error that pose other challenges. My workflow is meant to provide data that can be reviewed by more capable bat experts so I was going to see if there was acceptance over time defined by me as at least 50-75% of the observations getting to research grade.

That said, I understand full well it poses a problem with including photos that are not actually an observation of something. I have seen iNaturalist grow over its brief history and by all accounts it ranks a close second to the largest repository that includes photographs and sound so why can’t the sharp
minds in inat development sort this out?

If not, then it should be made clear that record entries like spectrograms are not to be submitted because in the absence of this policy, they are. It would be nice if there was an recognized alternative repository to send them to, made clear. Personally I keep my recordings and spectrograms on my home
computer so I could delete my iNaturalist copies and still have the data (which would be quite useless if kept locked away on my PC).

I would request that this issue be given serious review. I can wait, but not sure the bats can.
Thanks, I really do love everything about iNaturalist so please take this as some constructive feedback.

2 Likes

I think having a way to incorporate spectrograms would be useful, but that doesn’t change the fact that, currently, they are discouraged and cause issues for training the CV model - including them with observations could actually reduce the ability of the CV model to accurately identify the target species.

You asked for options for making spectrogram data available for observations, and staff did provide some alternative suggestions for how to include this data with observations in the link I quoted previously:

There are a lot of free image hosting services online that could be used to create links to add to your observation, so you could consider that.

In terms of adding functionality around spectrograms, @thomaseverest posted a link to a feature request to add spectrograms to sounds above:

you could also consider making a different feature request if that one doesn’t meet your needs/desires. Feature requests are the process by which many community-suggested features have been added to iNat. That said, there’s limited staff/development time for adding new features, so many are not added or take some time to do so (but you don’t know until you try!).

2 Likes

have you considered talking to organizations that are focused on bat conservation to see what they do to collect bat data? there might be a better repository out there for your bat data. or maybe you can convince one of those other organizations to contribute some time or skill or money to develop the necessary features within iNaturalist?

or maybe there’s some other standard that bat folks could agree on to make bat audio recordings more human scale. for example, maybe similar to the way a LUT works in image processing, you could have some sort of standard frequency shifting transformation for bat audio that would map the bat audio to frequency ranges that humans can perceive. if all bat folks used the same transformation, then anyone would be able to listen to the files using standard audio applications and equipment, and you would always be able to take an audio file generated from this transformation and reverse it to get back to the original audio, albeit with some loss of information.

But the same member of staff makes clear later in the same thread that there is no official iNat policy on excluding spectrograms. The views above appear to be personal ones: