My sat-nav, interestingly, manages to hold two pronunciations – “St Alb’ns” most of the time, but one or two junctions are “St All-bans”. Very disconcerting…

Has anyone ever built an IPA-to-speech-synthesiser system? It feels (on paper) like it ought to be practical, and it would be the core element to doing this – we could harvest the IPAs from Wikipedia or elsewhere.