search

WeidiXie / VGG-Speaker-Recognition

Utterance-level Aggregation For Speaker Recognition In The Wild
364 stars 97 forks source link

The initialization method #9

Closed DoubleKing0625 closed 5 years ago

DoubleKing0625 commented 5 years ago

Hi,

I saw that you used 'orthogonal' initialization for the whole network. Are there some special reason behind that?

WeidiXie commented 5 years ago

Hi, @DoubleKing0625

Not really, just my personal preference, though I never try other initialisation methods, but I believe this is not a key problem.

Best, Weidi

DoubleKing0625 commented 5 years ago

Ok, thanks~