The README states the code is dual-licensed MIT / Apache-2.0, but I couldn't find an explicit statement covering the pretrained weights.
Could you clarify whether the checkpoints under models/ fall under the same terms — specifically whether they may be redistributed, for example bundled with an application?
For context: I'm using DeepFilterNet3 in an open-source Windows noise-suppression tool, and I'd like to ship the model with it rather than make every user fetch it separately. I noticed Intel republish DFN2/DFN3 on HuggingFace marked MIT, which suggests the intent is permissive, but I'd rather not rely on a third party's reading of it.
Happy to open a PR adding a line to the README once you confirm either way.
Thanks for the project — it comfortably outperformed every alternative I measured on real recordings, including much newer models.
The README states the code is dual-licensed MIT / Apache-2.0, but I couldn't find an explicit statement covering the pretrained weights.
Could you clarify whether the checkpoints under
models/fall under the same terms — specifically whether they may be redistributed, for example bundled with an application?For context: I'm using DeepFilterNet3 in an open-source Windows noise-suppression tool, and I'd like to ship the model with it rather than make every user fetch it separately. I noticed Intel republish DFN2/DFN3 on HuggingFace marked MIT, which suggests the intent is permissive, but I'd rather not rely on a third party's reading of it.
Happy to open a PR adding a line to the README once you confirm either way.
Thanks for the project — it comfortably outperformed every alternative I measured on real recordings, including much newer models.