Although this would be feasible on its own (trimming audio on its own would just require listening to it repeatedly to confirm where you want to trim it), I think this would best be done in conjunction with adding spectrograms so you can see the sounds you’re trimming.
I record 1 minute sound files. I thought that some people may like to feel the ambience here for a minute or 2 at the spot where I was standing. At other times, I record for 30seconds. A sound file at 5 - 10 seconds sometimes do not convey enough information for identification.
As much as I do enjoy soundscape recordings, I don’t think iNaturalist is the place for this.
This can be true, although it’s often not. That said, I have no problem with a longer recording that aids in the identification. The thread above goes through some of the nuances here, but I think everyone would be on board with a long recording in order to get a more accurate/comfortable ID!
Sound recordings can be tricky because of the many non-target noises from aircraft to crying babies, never mind other birds, or even the observer’s footfall sounds and moving the phone around in their hands. I find I cannot use the in-app sound recorder very often, for this reason. Personally I strive to make recordings of just the target species.
Long ago I started using the Voice Memo app to make recordings. I can delete and repeat easily if inadequate. I can edit to clip off either end and reduce to just one species. I can duplicate the recording if there are multiple species, then reduce each to one species. I can record on my iPhone, then sync with my iPad to download the recording then upload to iNat. Seems to work well but is perhaps more cumbersome than many folks will have patience for. But its quality over quantity.
what’s this about transferring Merlin recordings directly to iNat? I’ve not found a way to do that. Also, the Merlin file is not compatible with iNat?
Will INaturalist one day coordinate with Merlin for identification of bird calls? What about Orthoptera and frogs? Seems a logical step that iNat would be working toward sound ID? @twane
I suppose a too-long audio recording with a lot of background noise and faint sound is analogous to an uncropped, blurry photo of a small subject organism. Both aren’t much fun to review.
I’m not that good with audio recordings and I use the iNat app to make them, although not very often. My process for a singing bird I want to record is to get as close as possible and then not move at all. Have the iNat app open and the audio recording ready to go. When the bird starts singing, start the recording and let it run for no more than a minute, often less (20-30 seconds) then stop it. If the bird continues singing, try one or more additional recordings, also around 30 seconds or so. I review my set of recordings before uploading them to iNat. Often there will be some background noise that will interfere with some recordings and you can discard those. If there are multiple birds of different species calling on a recording, I add notes to the observation IDing those and indicate which is the subject of the observation. I have not tried using Merlin to make the recording although I use it often to help confirm what I’m hearing before I activate the iNat app.
There is no way transfer directly, but it is relatively straightforward to save the Merlin output file (.wav) and then upload that into iNaturalist. WAV files are definitely compatible with iNat.
I sort of hope they don’t, since so many other teams have been doing this very successfully for a while. There’s no need to reinvent the wheel when there are excellent programs out there for identifying sounds of birds, bats, insects, and frogs.
Off-topic: where can apps/websites be found for ID’ing insect sounds from recordings? Would love to find one.
I’ve had people ID cicadas by song in just my iNat observations.
Same here, eg https://www.inaturalist.org/observations/142052820 It’s not always possible, but
FWIW, that’s pretty much what Merlin does, it’s trained on recognizing images of sonograms. Grant Van Horn, who helped us developer our CV model, also worked on Merlin. I can’t find the article at the moment (or it may be paywalled), but I believe they asked volunteers to go through sounds and spectrograms and mark exactly which parts of the spectrogram were the calls and then trained on just those parts of the spectrogram. Or something along those lines.
Fair, although when neither is high enough quality to ID on its own, sometimes you can combine subpar photos with subpar audio into a passable observation that can be IDed. I find those fun, although I know it’s subjective.
Some user (I forgot who??) has built and posted a browser extenstion that actually does this. It isn’t perfect yet, but it is pretty damn close for birds much of the time. Frogs…sometimes. Insects,…not really. It often picks up background species I hadn’t heard.
I have it and although I rarely use it as a source for an ID, it is kind of a fun extension for those of us who ID a lot of Audio Observations.
I can’t find the original forum discussion where it was announced, but here’s the link if anyone wants to try it.
In regard to identifying long/bad recordings:
As someone who does a lot of identifying of frog recordings (ok, I’m in third place…), these two issues (long recordings, lousy recordings) can be frustrating. But I’ve developed a workflow for identification that is tolerable. It is pretty difficult on a phone, but using a computer I can get through a lot of recordings pretty quickly.
I start off, of course, by clicking play in the identify screen. If the species they mention calls in the first 10 seconds of the recording or so, I agree. If they have only identified to a general category like “Frogs and Toads” or “Amphibians”, I will make a quick list of the species I hear in a comment with my ID being the most conspicuous species in the recording.
If the recording doesn’t provide the answer in the first 10-15 seconds or it is inaudible, I download the recording and open it up in Audacity. There I will clean it up by high pass filtering as needed, cutting out sections where there is talking/movement/cars passing/etc. and then amplifying the rest of the recording so I can see the spectrogram and pick out the target species (many species are easy to recognize on a spectrogram with experience). This sounds like a lot, but with a few keyboard shortcuts within Audacity, I can usually ID a frog call in a minute or less. I wish I could upload the cleaned up version for other reviewers to hear, but…it isn’t my record.
In my own phone recordings, I always download them later and clean them up before re-uploading them to iNat. Again, Audacity is awesome for this.
In regard to uploading recordings:
- I try to make sure my recording is between 10 and 30 seconds generally. That’s enough to give an opportunity for identification.
- I will filter/clean up recordings as my voucher recording. It isn’t exactly what I heard maybe, but the “cleaned up version” allows for easy ID.
- I may add a short (~30 second) ambience recording so the listener can hear the frog in its auditory context. I regard this as analogous to a “habitat shot”. In multiple cases, experts in that region have later pointed out species I missed and gotten me “recording lifers”!
- If I make a phone recording, after uploading the record from my phone, I will download the recording to my computer and trim and clean it up appropriately so that someone else can ID it. Sometimes my phone recording is just a placeholder for a voucher recording I upload later from my better gear.
Here is an explanation, co-written by Grant Van Horn, of how Merlin was trained for sound ID: https://www.macaulaylibrary.org/2021/06/22/behind-the-scenes-of-sound-id-in-merlin/. Sorry if this is getting a bit off-topic!
I remember for several years Cornell was asking volunteers to draw boxes around birds in photos, list which colours were most obvious, etc. I think Merlin does (or did) photo ID as well as audio but their photo ID never took off in the same way that iNat’s has done without any of that manual labelling. Photos do tend to more clearly have one subject in focus than audio though.
I avoided making more bird observations for a long time because birds are harder to photograph and I did not want to submit blurry photos. I challenged myself to make more bird observations last year and started playing with the sound ID in the iNaturalist app. Since I could not edit the clips, I’d make and discard clip after clip until I got one with the bird vocalizing within a reasonable time. Then I’d note the start time for the observation subject. Because it was cumbersome to make a decent sound observation, I only submitted a few.
I use Merlin but until reading this post, didn’t consider downloading those clips . . . I’d assume long sound clips were coming from observers like myself who mostly use the iNaturalist app to make observations. If that’s the case, adding the ability to trim a clip (in app) before submitting would be ideal. A pop-up asking users to trim without it being an available feature would be off-putting.
Learning I can make & upload external .wav files to get around the (current) limitations of the iNaturalist app recording feature is a game changer! It’s kind of a duh! moment to be honest.
I’ve been using iNaturalist for almost 5 years and I’m still learning, still consider myself a casual, generalist observer. So I wonder about the learning curve and how much of this issue is the disconnect of causal app users from the experienced community in the full site/forums? I’ve learned so much by randomly browsing the forums and from the comments others have left on my observations, but often feel like I’ve barely scratched the surface! Hearing what the experienced identifiers need and their strategies has helped me make better observations. Even though I can’t answer your question as a bird identifier, I wanted to thank you and the commenters as an observer because posts like this help me learn & improve my skills.