License for the pretrained models?

[hbredin]

Thanks for the great work!

I understand that the code is licensed under MIT.

However, what about the pretrained models? Do they allow commercial use?

I am asking because I would like to possibly integrate them in pyannote OSS speaker diarization pipeline.

[vanIvan]

For the models weights we are applying same license, as wespeaker models checkpoints have, as training pipeline with some modifications + data are the same:

The pretrained model in WeNet follows the license of it's corresponding dataset. For example, the pretrained model on VoxCeleb follows Creative Commons Attribution 4.0 International License., since it is used as license of the VoxCeleb dataset, see https://mm.kaist.ac.kr/datasets/voxceleb/.

Please do not forget to star our repo, if you find it useful.

We are also looking at expanding our research onto diarization, we also would like to build diarization on pyannote exploiting our models - probably we can work here together? The straightforward way is to incorporate ReDimNet model as speaker embedder model, however we believe there is a way on updating local speaker segmentation model. We asked in discussions for hyperparameters / recipes for training custom speaker segmentation model using pyannote, however your colleague only directed us to tutorials, which don't contain necessary hyperparams / data configs for training speaker segmentation model from scratch, probably you could share this info with us?

[hbredin]

Thanks for your answer. Feel free to drop me an email about possible collaboration.