是在sv ai gen3更新后出现这种情况，多个ai歌手的品质渲染均有杂音，提升气声后杂音尤其明显
sv gen3 弦卷真纪ai问题反馈.zip (6.1 MB)
seems like this
Currently, Lite & high-speed mode did not being affected by the Tension parameter “problem.”
The real problem is, it may need to retraining to fix that or not.
It will need to be fixed by the next-gen model since every retrain time costs just the same.
I’m not Kanru Hua.
I don’t know.
SynthV engine can not “generate” more high tension than real record sample,
So you need to record it as a “high tension” sample and let engine “soften” it.
This basically what occurred when recording Haiyi that her old library has quite a lot of soft samples to making her singing style characteristic, but her SynthV library records high-tension samples making her default character not quite the same as the old one.
This "should not affect AI, " but I think this “problem” shows us not really like that.
At SynthV Standard, this problem is usually addressed by the default value set in the library.
Because it did not change a lot at SV-std-era, we did not face that problem earlier.
But when we assume it is affected, then the default value should be changed when engine update can affect tension parameter.
This makes you need to review the default setting of the library when the AI engine gets generation forward.
Basically library side did not have the time to review their library when they got the engine update. It is just as fresh as from the oven.
I think you write at here still help to get it to fix.