The generalization abality

[ABexit]

The generalization of RawNet2 is poor?
I trained RawNet2 in AISHELL dataset with 340 speaker and tested in trail.txt with 8w pairs bulit by another 40 speaker of AISHELL, and the final eer is 3.46%. But when tested in 40 speaker of VCTK dataset with 8w pairs, the eer got 32.71%. Do you know why? Thanks.

[Jungjee]

Hi, it's not easy for me to judge the extent of generalization.
I don't know how much difference AISHELL and VCTK datasets have.
However, normally, I would not expect EER to increase over 30%.
One example would be cross lingual experiments (train: English, test: Korean) which is not published.
In this case, EER was somewhere between 5~7%.

Hope this helps :)