I made some bat audio recordings with an ultrasonic microphone. Of course humans can’t actually hear the bat noises in the recordings, so I was wondering if I should slow down the recordings before I upload them. If I slow them to 1/10th speed, they are audible for humans. But I’m also wondering if that might throw off some future AI audio identification feature in iNaturalist. Should I upload them raw or slowed down? Opinions?
Here’s one slowed down to 1/10th speed: https://www.inaturalist.org/observations/66030666
Oh…! It sounds cute! So cool that you shared this opportunity to hear a bat vocalize.
At this point, I’m not sure there are any standards or best practices for audio uploads. We’ve discussed with iNat staff the possibility of having better spectrogram support (and therefore, image recognition support), but it is a whole new feature and I’m not holding my breath that we’ll see anything soon.
Upload both, or either. Audio observations make up a scant fraction of all iNat observations, and I love hearing about people contributing anything at all!
I think 10x slowed down should be the best. Anyway, any AI should be able to guess the difference between 1x or 10x on the basis of sample rate
If you ever have a question of “should I do something that will throw off the AI but make it easier for other humans to interact with my observation?” the answer is “do what you want, don’t worry about the AI.” iNaturalist is about helping you and others connect with the natural world. We’re not actually here to help feed clean data to an algorithm. The AI is there to be a tool for us, we’re not here to train the AI.
Thanks for the advice all! I uploaded a couple more files at 1/10th speed:
https://www.inaturalist.org/observations/66054745 (probably Hoary Bat)
https://www.inaturalist.org/observations/66055763 (probably Silver-haired Bat)
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.