Speaker-Diarization icon indicating copy to clipboard operation
Speaker-Diarization copied to clipboard

Published 20 hours ago • verified publisher icon taylorlu

Can speaker model be reproduce?

Open zh794390558 opened this issue 4 years ago • 4 comments

Can you use https://github.com/WeidiXie/VGG-Speaker-Recognition or your re-implementation (https://github.com/taylorlu/ghostvlad-speaker) to reproduce pretrained speaker model ? If yes, how?

zh794390558 avatar Jul 22 '20 02:07 zh794390558

In your first href.It's to produce weight.h5. 1595398941926

jeremy110 avatar Jul 22 '20 06:07 jeremy110

Does the batch size must be 160 with 2 gpus，since when i using this params, OOM occurs？ Or can i train with 4gpus and small batch size which fit to gpu memory?

zh794390558 avatar Jul 22 '20 08:07 zh794390558

You can directly reduce your batch size and increase the epochs.

jeremy110 avatar Jul 22 '20 09:07 jeremy110

Thanks, I will try with this cmd.

zh794390558 avatar Jul 23 '20 02:07 zh794390558