Anyone working with a custom classifier?

[wcornwell]

Hi all,

I'm wondering if anyone else is using a custom classifier?

I've gotten much better precision and recall with a regional classifier trained with the birdnet_analyzer train approach. Also better performance than Perch v2.

Can also add local non-bird animals pretty easily also. And using local sounds as (non-event) noise massively improves performance.

Just interested in trading notes and approaches.

Cheers,
Will

[hugokoopmans]

i would love to exchange idears, we are planning to look into frog vocalizations at the moment...

[wcornwell]

hi @hugokoopmans, sure happy to chat. it's just a question of the training data on the right species of frogs to get it to work. And there is a strategic decision about whether to do frogs only or frog + birds or frogs + birds + crickets.

@Nachtzuster the reason I posted here was that it was slightly hacky to get a custom classifier to work with all the cool features that you (and others) built in this repo. Was wondering if it would be worth making swapping classifiers a bit easier on the RPIs.

[Nachtzuster]

@Nachtzuster the reason I posted here was that it was slightly hacky to get a custom classifier to work with all the cool features that you (and others) built in this repo. Was wondering if it would be worth making swapping classifiers a bit easier on the RPIs.

suggestions?