Description of need:
Audio-only observations often remain stuck in ‘unknown’, due to a limited amount of audio identifiers as well as the issues with audio workflow highlighted by @taitsougstad. However, as far as I have seen and can reason, audio recordings are always of animal sounds (in a rare case of ‘inappropriate use of sounds’ it would be a recording of humans, i.e. also an animal).
Feature request details:
iNat could default all audio recordings that are not assigned any identification by the observer to ‘Kingdom animalia’ (in a similar way that it now automatically assigns certain observations to ‘not wild/cultivated’). This should make it more likely for these recordings to end up in the ears of a capable identifier and would reduce the number of ‘unknowns’.
It’s rare, but I have heard a couple of observations with audio recordings of plants, capturing the sounds of pods exploding to release seeds or the distinctive rattle of dry fruit with seeds. Both of those also included pictures, though.
Good point overall. While there are edge cases like plant sounds, the vast majority of audio-only observations are animals, and defaulting them to Animalia would probably reduce a lot of friction and “unknown” backlog. Maybe an optional prompt or soft default (with an easy override) could balance accuracy with discoverability.
Seems… unnecessary. Are people identifying audio by filtering “audio only, animals only?” Or are people who identify unknowns sick of identifying audio? Whether it’s left unknown or specified to a kingdom doesn’t seem to make a lot of difference for quality in most cases and it doesn’t seem like it would help get it to the specialists any faster (in fact it might be slower because I’d figure there’s more “unknown” IDers than “rank: kingdom = animalia” IDers)
Since relatively few people search for “Unknowns” and people searching for animal observations just wouldn’t, this proposal might be useful. I think we’d probably be able to correct the few plants that have audio files pretty easily. Not a high priority for me, but low priority on the positive side.
I haven’t tried to upload a recording so I don’t know. However, with all the info on iNat, is it possible for the site/app to make suggestions as it does with images? Merlin as able to ID calls. Why not iNat?
Merlin technically trains on the spectrogram images of the recordings rather than the audio itself. This works with the eBird/Macaulay Library audio database because a spectrogram is automatically generated for all uploaded audio. iNat currently doesn’t produce spectrogram images so CV training on them isn’t an option at this point, although it’s theoretically a possibility for the future.
I checked a few of the stranger ones to supposedly have sound, and it seems sometimes people include audio recordings of themselves saying what they have found. I don’t think that’s the intention of audio recordings, but that’s a story for another day.
Thanks everyone, for weighing in with thoughtful responses!
To be clear, I’m aware of the possibilities to filter for/out audio files. This is not so much a request for my personal convenience, but was meant to
I do believe a measure like this could make a real impact in terms of quality/access to experts. As I primarily identify plants, I usually don’t look at unknowns, and when I do there is little I can contribute to most audio-only observations, so getting them directly to animal-only identifiers who might never look at unknowns seems useful. Essentially, @sedgequeen captures my thinking quite succinctly:
However, I am swayed by @martyndrabik and @phma ‘s search and examples: there is legitimate use for audio-only plant and fungi observations and arguably we’re better off not defaulting to a wrong ID.
I suppose the question then becomes what tolerance do we have for this type of error (audio-only non-animal obs ID’d as animalia). We currently have 1,400,000+ animal audio recordings versus 1,300 non-animal audio observations, which would suggest this error rate might occur somewhere around 1 per 1000 defaulted IDs. That’s 10 wrong IDs to move ~10,000 audio observations currently in unknown (about 2% of all unknowns) to Kingdom-level.
I don’t think the issue is your ‘10 wrong IDs to move’. I think people who want to ID audio, filter for that. Moving an obs from Unknown is not an end in itself. It still needs an active identifier. Dumping 1.4 million audio recordings in animal - does not achieve anything.
I mostly ID plants, but I do look at Unknowns thoughtfully. We each approach identifying differently.
I’m a little skeptical about both how many people are IDing at kingdom Animalia specifically (rather than not filtering taxonomically at all or filtering to finer ranks of Animalia) and what proportion of those are in turn willing/able to ID most sound observations. As stated elsewhere in this thread I think
is almost entirely due to the lack of sound IDers. I guess taken literally moving every unknown sound ID to Animalia would “solve” this problem, but I don’t think there’s much of a difference in a rank unknown sound observation and a rank Animalia sound observation both in terms of likelihood of getting an ID and in general quality of the ID. Actually if I had to guess I would suspect an unknown sound observation is slightly more likely to get an ID than a rank Animalia one because of the large number of people specifically IDing unknowns.