Idea about Hifisampler in OPUM

[Electric-wind]

I think we can use the mel features from the nsf-hifigan model for concatenation instead of Wavtool, which not only improves the results but also reduces rendering overhead on mobile devices.
Example：https://github.com/xiaobaijunya/HiFiUTAU-engine

[MysticILD]

Great idea! I was actually trying to fix the accuracy issues of HifiSampler, so I'll check this out!