Noise, negative samples & append setting

[nishantnnb]

Hello @kahst,

I have a few questions regarding training custom model - mainly to reduce false detection.

Question 1)
Currently I am facing a problem: With Birdnet model V2.4, Dog howling (+ barking) is getting detected as owl.

How to tackle this?
Collect these dog sounds under 'Noise' folder (class).
and train a new model with Append setting.
But the issue #284 states that the system doesn't accept a new class with same name (Noise) which is already present in the base model.

Question 2)
Can we use "negative samples for classes by prefixing the folder names" with Append setting while training a custom model?
e.g.
Birdnet V2.4 already contains a class for:
'mottled wood-owl_Strix ocellata'

I will create a new folder only for '-mottled wood-owl_Strix ocellata' (negative folder) and put all dog sounds (and or other negative sounds) in it.
And train a new custom model with Append setting.

Is this possible to do so? And will the model not detect (or reduce detection of) dog sounds as Mottled wood-owl?

Question 3)
I have trained a model for say 100 birds with the bird sounds from my area. I want to detect only specific birds (e.g. 5 types of owls) with this custom model. How to achieve this?
I could use custom species list in Species selection, but it would use the standard Birdnet 2.4 model and not the custom model trained by me.
Is it possible to provide the option (as an enhancement) to choose custom species list with custom classifier ?

Regards,
Nishant