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

How do you apply AS-NORM? #12

Closed zabir-nabil closed 2 years ago

zabir-nabil commented 2 years ago

Hi, thanks for sharing your code. You say the best performance is achieved with AS-NORM. Can you share how you apply the AS-NORM with which set?

TaoRuijie commented 2 years ago

Hi, AS-NORM is performing in the VoxCeleb2 training set. To each test speaker, we find 400 similar speakers from the training set. So that build a more robust score matrix.

Sorry, I am on holiday and reassembling my server. So I may not update it recently.

I did that based on this paper, you can check it as the reference: Analysis of Score Normalization in Multilingual Speaker Recognition. Interspeech 2017.

Thanks for your understanding.

JJun-Guo commented 1 year ago

Hi, thanks for sharing your code. You say the best performance is achieved with AS-NORM. Can you share how you apply the AS-NORM with which set?

Hi, Do you try AS-NORM with the VoxCeleb2 training set ?