search

TaoRuijie / ECAPA-TDNN

Unofficial reimplementation of ECAPA-TDNN for speaker recognition (EER=0.86 for Vox1_O when train only in Vox2)
MIT License
581 stars 111 forks source link

test performance #75

Closed wwyl2000 closed 5 months ago

wwyl2000 commented 5 months ago

Hi, Thanks for your great work! How to test performance?

TaoRuijie commented 5 months ago

If your testing data is in the same domain of VoxCeleb1 or 2, you can use the same threshold. The threshold comes from ComputeErrorRates function. (can also come from ComputeMinDcf is you focus on mindcf)

If you are using other dataset, a developmenet set is required to decide this threshold.

wwyl2000 commented 5 months ago

Thanks for your help!