search

TimoBolkart / voca

This codebase demonstrates how to synthesize realistic 3D character animations given an arbitrary speech signal and a static character mesh.
https://voca.is.tue.mpg.de/en
1.14k stars 271 forks source link

question about the learning rate decay #102

Closed brooks0519 closed 1 year ago

brooks0519 commented 2 years ago

Hello, Thanks for your great work, recently, I have do some work to replace deepspeech0.1.0 to adapt live audio streaming inference. So I retrain the voca model, but I found the learning rate is always 0.0001, I want to know have you ever use decay learning rate? and the effect of different learning rate? thanks again

TimoBolkart commented 2 years ago

No, we did not decay the learning rate and we did not investigate the effect of different learning rates. If you find some better training strategy, I would be happy to learn about it.