akfheaven issues

Results 3 issues of


                                            akfheaven

比dlib demo还慢，求优化。

Can ffmpeg work this way ?

I've tried normal speech dataset and generated very natual voice. But how about training with emotional dataset? any one have a try?