Thanks @olivier2222 I'll look it up, but I think Claude already used it. I've basically reimplemented narrator.device 33.2 in TypeScript (github.com/bitplane/narrator-ts) by measuring and reimplementing the code side of things, but the specific translator (letters to phenomes) and narrator (what they sound like) tables are copyrighted data that we don't have a license for. So Claude found this Naval Research Lab doc that early Amiga translator was based on (PD), and had it make its own narrator that's kinda crappy sounding but similar-ish to SoftVoice's, then iterated on it listening to them. I think tuning to sound more like the Amiga without violating copyright is the right call here, but it'll take a fair bit of work