Just to avoid thread hijacking, I’m moving the discussion here.
Here’s what I’ve learned from analyzing de-essers. A “TH” sound is basically a “SS” sound with a bit of a low pass filter. This is why De-essers tend to give a lisp, because they are usually a high frequency compressor, which turns down the SS by filtering out high freqs, but the natural consequence is that it turns into a “TH”
If you turn down the SS, you have to also turn down the low frequencies. Which means you need a solid detector that can know with high level of certainty that the sound it is hearing is an SS sound. This is actually not that hard, as I use Big Beautiful Door to do this. But I think the detector could be a lot more robust with a bit more thought put into it.