I may have worded this wrong in the frenzy of discussion yesterday.
What’s going on with VoiceSeparator wasn’t as much a case of tool A is worse than tool B, and therefore I hate tool A. What I did in my test more or less mirrored exactly what was done in the live stream and also demonstrated in the separate Nuendo DX feature highlight, except my results were different.
So it felt more like a bug, not a feature that behaves differently. Bugs are not uncommon in major releases, and it was just disappointing that I couldn’t replace my other tool in place with this new tool, when following exactly what the release demo suggested should work.
After further testing it’s become clear that this may indeed be a ‘bug’. The problem only occurs in the first few frames when there are no handles. It appears the tool needs a few frames of tuning time before it works properly. That’s not something the other tools require, and was not evident in the demo as there they also lowered the slider to or near infinity, but they did so while playback was already going on, not at the start of an event.
It seems like this should be looked at as a bug or improvement, so people cutting this in too close to an event start may find it frustrating. And by all appearance the degradation may exist at lower strength settings too, just not be as apparent and may be missed, which would still make it a problem for discerning operators.
The fix most likely may involve some sort of look ahead to allow the algorithms learning/adaptive logic to see what’s coming when it’s processing the very first frames of an event. That may require an unfortunate latency but that is the nature of these algorithms.
And yes, I have all these other tools, which is why I was able to compare. And as we all know some of these tools work better with some material than others, so I was excited to have one more at my disposal, and somewhat disappointed that a headline feature appears to have a bug in it.