MCTT

it works,thanks very much

I'm confused too @hannarud

me too. and i found the spectrum of vocals.wav and original audio are the same except the volume lower. i wonder waht's the problem.

> Is this training creating a model from zero or finetuning the May model? I trained model from zero on my own data，use --au45 but it can't blink.

这种抖动看着像拍摄时人物抖动过快或太大了吧