Relicense as Apache 2.0

[Bengt]

I would like to combine this project's code with the code from birdnetlib. Since birdnetlib is licensed under the terms of the GPLv3 and this project is license as CC-NC-BY-SA 4.0, that requires licensing this code under the same terms. Please resolve this licensing conflict by relicensing this project under the same license as birdnetlib.

Note that relicensing birdnetlib to CC-NC-BY-SA 4.0 has been discarded, here: joeweiss/birdnetlib#72

[DD4WH]

There are some really complex open questions here:

• The BirdNET models have been trained with data available under different licenses (from XenoCanto and Macaulay) --> how should one license BirdNET Analyzer then ?
• And: Do the python code and the model itself need different licenses ?
• Is the "non-commercial use" part of the existing license compatible with research projects funded 100% by a commercial company ? Is the license compatible with research projects that are publicly funded ? Can BirdNET-Analyzer be used for teaching at privately funded universities ? (which is often interpreted as commercial use) Can it be used in teaching at publicly funded universities ?

I have no answers to these questions. But my gut feeling would be that we should leave the license question open/vague and instead focus our efforts on working with and improving BirdNET as one of the top research tools for bird (and other animals´!) sound classification. I also guess / feel that this could be in the spirit of the maker(s) of BirdNET.

Another reason is that I would not really want to have lawyers decide which kind of ornithological research strategy we are allowed to carry out . . . we should be very happy that the makers of BirdNET provide us with such an excellent freely available tool! Thanks Stefan et al. !

[Bengt]

Hi, @DD4WH!

I am not a lawyer, but:

• My understanding is that models trained on licensed data do not need to be licensed alike the data. However, one would need a license / sufficient rights to use the data for training in the first place. In practice, this is not enforceable, since the training usually happens in private, so data piracy is unfortunately commonplace.
• The Python code should have a license which applies to and makes sense for code, preferably an OSS-approved one. I feel like the models can just be released and licensed "with the code" as one would do with any other non-code artifact like configuration files, images or documentation.
• Additional challenge for the lawyers: I am working in research at a gGmbH, which means it is privately owned, state subsidized, project funded, non-profit with >1000 employees, some of which are professors at various private and public universities. Are we commercial or not?

I am not requesting a license change for the sake of it, but I would actually like to integrate the birdnetlib code with my fork of BirdNET-Analyzer fork. That way, one would have a project that can infer, train and create distributables, all of which are essential functionality for some of my use cases. I feel like the current licensing situation actively prevents me from doing so. Is that intended?

Sure, the lawyers should only advise. In my experience, there are issues to be discussed with them, and this discussion also shows that there is a lot of uncertainty revolving around the licensing. However, things become simple when you realize that researchers like Stefan Kahl are publishing in line with a research strategy which takes precedence over legal concerns.

[Bengt]

Note that birdnetlib is now Apache 2.0, I am changing this my request to following them:

joeweiss/birdnetlib#77