PlayVoice/VI-SVS

full song with dnn pitch

MaxMax2016 opened this issue · 3 comments

pitch is no easy to predict

nn_pit_song.mp4

Oh wow ! , that's indeed way more better and smoother than the other one , so what is missing now for the implementation to be fully done ? can you put the missing stuff in like a code summary , as sometimes I think we are lost in the translation from Chinese to English :) , so i see if i can help you or get help. I saw this today, not sure if that would work for SVS tho but here it is https://github.com/yl4579/PitchExtractor also that https://github.com/r9y9/pyreaper

pitch is no easy to predict

nn_pit_song.mp4

您好,请问 这个nn版本 pitch 的infer,是采用哪个infer脚本,是“pyton vsinging_song.py”吗?
如果不是,想要达到您提供的这个效果,需要怎么做?望不吝赐教!

@Lukelluke 简单的讲,就是使用帧级别的phone,midi通过transformer预测帧级别的pitch,相关pitch预测可以参考
https://github.com/MoonInTheRiver/DiffSinger/blob/master/modules/fastspeech/pe.py
https://github.com/yl4579/PitchExtractor