Currently, you can only try the new features (like the “enhancement” mode and Vibrato Modulation) in the latest version of each voicebank. Sadly, I am not pleased with how some of the voices sound with the latest update, so I’ve been forced to go back to their previous version, which makes the newer features unavailable. Can you please make the new features available to all previous versions of the voicebanks?
Doesn’t that defeat the purpose of a “version” update for the voice databases then? Because the previous versions were not compatible with the features prior. The update is what makes them functional.
If that’s the case, fix the voicebanks so that they sound as good as before. What’s the point of adding new features if it’s gonna mess up the voices?
There aren’t any situations where they intentionally make the voices sound worse. RLHF is clearly intended to improve the automated behavior, not just add a slider for the sake of it.
Generally, most people seem to feel 1.9.0 improved the generated pitch patterns, in addition to giving much better user control with note-level settings instead of project-level ones that many of us considered to be borderline useless.
As for 1.10.0b1, it’s a beta. The whole point of the beta period is to catch these issues and fix them before release, so that things do sound better than before.
If you aren’t happy with the new sound, send Dreamtonics an email, or they might not be able to fix it before the final update.
Of course, the onus of testing shouldn’t be so heavily on the users’ shoulders, but that’s a separate issue…
Migi is also correct though, you can’t use RLHF enhancement on a pitch model that doesn’t contain any of the feedback analysis that the feature relies on. Similarly, you can’t use vibrato modulation on a pitch model that does not distinguish between vibrato and non-vibrato patterns.
The updates to the models are a prerequisite to feature compatibility in the first place.