I am not going to get into a technical argument. But it makes no sense to say that you don't process text that gets passed to a synth prior to its being passed to it. None of the dictionaries, to my knowledge, post process text after it goes to the synth, but are used to decide what, exactly, gets passed to the synth.
I don't even know what is referenced by "translation layer" and that may be the problem. But the final part of the process has to be handing off that which is to be spoken to the synthesizer that speaks it. All changes to same must occur prior to that step.
Brian - Windows 10 Pro, 64-Bit, Version 1903, Build 18362
The color of truth is grey.
~ André Gide