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
594 stars 113 forks source link

Best EER and the channels used #79

Closed wwyl2000 closed 5 months ago

wwyl2000 commented 6 months ago

Hi, Thanks again for your work. My best EER is only about 1.03%, and i need more epochs to get this.

Thanks, Willy

TaoRuijie commented 5 months ago

I did not set the seed so the results might be changed. usually I can get around 1.00 EER.

From my experiments I find the device difference can also lead to some slight difference. My best performance is 0.94, the worst is 1.03 (same to u)

1024 is a good choice, it can lead to the best performance.

wwyl2000 commented 5 months ago

Thanks Ruijie!