GE2E params added into optimizer and gradient scaling adopted - Githubissues

yistLin / dvector

Speaker embedding (d-vector) trained with GE2E loss

272 stars 46 forks source link

GE2E params added into optimizer and gradient scaling adopted #2

Closed cyhuang-tw closed 3 years ago

cyhuang-tw commented 3 years ago

I made GE2E parameters optimized and used gradient scaling during training.