how to finetune a pretrained model with new speakers

TaoRuijie / ECAPA-TDNN

Unofficial reimplementation of ECAPA-TDNN for speaker recognition (EER=0.86 for Vox1_O when train only in Vox2)

MIT License

563 stars 110 forks source link

how to finetune a pretrained model with new speakers #73

Open wwyl2000 opened 4 months ago

wwyl2000 commented 4 months ago

I have a question about the continuous training: How to finetune the model?

Thanks!

TaoRuijie commented 4 months ago

I am not very sure since I have not tried this setting.

From my opinion, since X >> N, finetuning with N speakers with the additional data only with a smaller learning rate maybe is the best. Fineting with (X+N) speakers with all data only with a smaller learning rate is also a suitable solution.