Is there a MTFG equivalent that works for lewd audio (not music based)?

I’m talking about videos like this: Backrooms Girl Entity

Comics or videos that otherwise lack movement, but you can script based on sound effects (not music). Not too hard to script, but I feel like something akin to MTFG for audio like this should be a thing. Like, “set a 0 point each time this sound plays.” All the audio based funscript software I’ve seen on here all seem to be geared towards pmvs and beat-based videos.

Not that I know of. I imagine for the hardcore option, one could use BS-RoFormer (Music Source Separation With Band-split RoPe Transformer) to train a model to perform stem separation with categories such as moans, speech with different moods, certain skin against skin noises etc.

An old example from Sony:

From the output of such a model you could simply take each stem category and assign script movement patterns to timestamps of peak amplitudes of the waveforms’ transients and modify each pattern instance duration to match the tail following each such transient.

You’d need a lot of high quality audio samples of each relevant category to train the model and the work to organize, label and mix the data as well as performing the training would be significant.
(Also, do respect the licenses and terms of available audio samples. I recently mentioned a small dataset of NSFW sounds and vocals, but that data is not permissible for training AI models on. Do not go against the wishes of people who provide valuable assets to the public by abusing their rights.)

Not sure, maybe I totally just don’t get what you are looking for, but it sounds more like a volume detection?

I’ve played around with that, but never wasn’t happy with the results, it sounds good in theory, I was thinking about vibrator and moaning, the louder she moans, the more it vibrates, but again, the play-tests doesn’t really felt good.

Anyway, something like this maybe:

Mi,11_39_17_.funscript (18.6 KB)

Done to the linked example clip, not 0 at sounds plays, the opposite, maybe you’d need to invert?
And you’d need to simplify and speed-limit this thing, too.

They’re asking for the script to depend on specific sounds’ volume, not the overall audio track. Otherwise yeah, you’d either do a cleaned up version of that or a threshold triggered action/pattern.