Speaker diarization - which layers are responsible?

#26
by deathknight0 - opened

Just wondering which layers are responsible for speaker diarization in the model? I want to fine tune it for domain specific vocab only, and would like to leave the layers responsible for speaker diarization alone . All my samples are single speaker and I do not want to introduce catastrophic forgetting for diarization tasks.

Thanks in advance!

Sign up or log in to comment