search

Jungjee / RawNet

Official repository for RawNet, RawNet2, and RawNet3
MIT License
357 stars 55 forks source link

Overfitting on VoxCeleb #28

Closed Spectra456 closed 1 year ago

Spectra456 commented 2 years ago

Hi. I tried pre-trained RawNet on VoxCeleb and it's works better than ECAPA-TDNN pre-trained on VoxCeleb. But when i'm switching to another dataset on English(without training) or other languages it's works worst than ECAPA-TDNN, usually 2-4 times (EER). Do you have any ideas why it's happing? Thanks!

Jungjee commented 2 years ago

Hi @Spectra456. The easiest thing to say would be that RawNet3 (I assume you used this one, since ECAPA-TDNN outperforms RawNet2) is somehow overfitted to VoxCeleb domain. However, it would be better to be cautious about making any claims at this point.

May I request for some additional information? such as the datasets you've tried, their corresponding EERs, minDCFs, ...

Jungjee commented 1 year ago

Please reopen this issue when you have further updates.