Licence clarification: are the pre-trained models covered by the MIT licence?

Hello,

I'm doing licence diligence before choosing a source-separation model, and I'd like to
confirm my reading of Spleeter's licence for the pre-trained models, because two of your
own publications appear to say different things.

The JOSS paper states, under *Distribution*:

> "Spleeter source code and pre-trained models are available on github and distributed under
> a MIT license."

The repository README states:

> "The code of Spleeter is MIT-licensed."

The first sentence covers the pre-trained models explicitly; the second mentions only the
code. The LICENSE file is a standard MIT text and does not distinguish between the two.

My question is narrow: are the pre-trained model weights distributed as release assets of
this repository covered by that MIT licence, including for commercial use?

I ask because the paper also states the models were trained on Deezer's internal "Bean"
dataset of 24,097 tracks rather than on a research-only corpus, which is why the question
seems answerable at all — for most published separation checkpoints it is not.

I intend to rely on the answer in a commercial context, so I would rather ask than assume. A
one-line answer either way is enough, and a public one here would help everyone with the
same question. I'm not asking for any additional grant or a custom licence — only for
confirmation of what the existing published licence covers.

Thank you for Spleeter, and for publishing the training details; that transparency is the
reason this question has an answer at all.

Best regards,
А head of evaluating Spleeter for a commercial product