I'm off to bed but my computer cranks on. It's using #BirdNet to ID all the bird songs recorded by our garden's #AudioMoth from the first half of the week, while an R script is hooked into AWS to transcribe the 772 geotagged audio notes I made today on my monthly 12 km run up from the city through Victoria Park and around the native forest on Sugarloaf peak.

Some parts of our crazy Big Tech timeline I'm OK with. 😄

#wildcounts #EcologicalMonitoring #nz

@joncounts Do you have the code for this available? And do you use BirdNet's models - aren't they poor for NZ?
@richlitt Most of the birds we get in our garden are not NZ natives (lots of introduced European species) so BirdNet should be well equipped to ID those. It is detecting natives like fantails and bellbirds well enough to show sensible patterns, although it’s mixing bellbirds with tūī often. My next step is to compare BirdNet’s detections with my manual counts to see just how good it is.
@joncounts it's pretty bad at Whitehead and Riroriro, too, I've noticed. I was looking at some yesterday in Zealandia while recording and only Fantail was coming up...
@richlitt Also, I’m happy to share the R code. I’d need to annotate it better before putting it on GitHub or GitLab. I should do that soon! I’m happy to email it to you earlier if you want to try it.

@joncounts If you throw it as it is on GitHub or GitLab, I can work on annotating it. I think I've enough familiarity with birdnet to know how it works.

@tungite Do you know if birdnet has a way of adding your own models for bird recordings? We may need that in NZ...

@richlitt To clarify, my R code is for using AWS Transcribe to bulk transcribe all my audio notes that I make as I move about the city. For BirdNet, I'm just using the free computer app to batch process my folders of audio files.

AWS Transcribe with spoken words is still good rather than excellent for my field recordings, and human voice transcriptions are a much bigger business than bird song. I expect it will take many more years for BirdNet to be as good as trained birders at bird songs.

@joncounts Honestly, I'm more interested i the AWS Transcribe! I've been meaning to build an app to log birds while driving by just talking to Siri. This sounds similar?
@richlitt Absolutely! I've found that by far the quickest, most convenient, and data rich way to make field observations of nature is to talk and keep moving. I tend to restrict my counting to when I'm a passenger in cars, for safety, but spend most of my time on a bike or on foot and so talking works great. I'll see if I can get my code shared in the next few days so you can see how I do it.