The SC-01 was decapped and its ROM extracted back in 2007, but the SC-02 has different phoneme data. MAME and AppleWin both just map SC-02 phonemes to SC-01, which doesn't produce accurate speech.

There's a high-res die shot of the SSI-263P on visual6502.org (Greg James / Christian Sattler): http://www.visual6502.org/images/pages/Silicon_Systems_SSI_263P_die_shots.html

I don't know enough about ROM extraction to tell whether that image is sufficient or if delayering would be needed. Looking for anyone who's done this kind of work or can point me in the right direction. I have chips and can cover costs.