Detecting recordings containing human voices?

[esfel]

Hi,
I was wondering whether it's possible to detect human voices with BirdNET: I'd like to store my audio recordings publicly, but I know that (very rarely), human voices are audible on those (recordings are from forests, deep in, but well, I guess some like to wander off at times). For data protection reasons I'd like to not publicly upload such recordings. As it's impossible to listen to all recordings myself, I would need BirdNET to tell me in which recording human voices are audible. I remember that, years ago, the BirdNET phone app detected (correctly) "Homo sapiens sapiens" on one of my recordings one time, but now I can't find Homo spec in the Labels list of the current model version 2.4 (https://github.com/birdnet-team/BirdNET-Analyzer/tree/main/birdnet_analyzer/labels/V2.4), so I guess creating a custom species list with just Homo sapiens sapiens and then running BirdNET on all my recordings with this list is not an option. Did anyone ever come up with a solution for this?
Many thanks in advance!
Esther

[tphakala]

The BirdNET model contains human vocal detection. The following labels are included in the v2.4 classifier:

Human non-vocal_Human non-vocal
Human vocal_Human vocal
Human whistle_Human whistle

However, the reliability of human vocal detection is another topic. The classifier in v2.4 isn't very good at this, but it is possible to improve human speech detection by appending additional speech training data to the default classifier.

[esfel]

Thank you so much!! Can't believe I've overlooked that in the list. Good to know also that the classifier is currently not very good at this.
Cheers, Esther