According to the Evaluation Plan, only the VoxCeleb2 dev set which contains 1,092,009 utterances from 5,994 speakers can be used to train the speaker model. Is this also applied to the training of PLDA?
We have scheduled a important maintenance operation for the Codalab public instance (http://competitions.codalab.org) for Saturday, September 8th.
The migration will start at approximately 9AM PST and should hopefully be completed by approximately 1PM PST that same day.
During that time the server will be down.