The NEST layer outputs were meant to be used as speaker embeddings, but they didn't function as such.
I also added variants for StreamingSortformer v2 (the old one was v2.1), which performed better in DIHARD III and may work better outside of meeting environments (unconfirmed).
- I think I fixed the model (77630916caa1977b8fad9697e617fbd7a995ce01)
- Added encoder outputs to the models for future speaker cache swapping (10281560623ebe5d72c3b2f398a6a7c61a418262)
Co-authored-by: Benjamin Lee <GradientDescent2718@users.noreply.huggingface.co>
