Talk:SmileBASIC (Petit Computer 3)/@comment-26924640-20150907162603/@comment-24739388-20150909202648

English has a lot of sounds, and simply handling the text alone is difficult for a machine to handle. I wouldn't exactly call it "very feasible", more like "doable but difficult and probably inefficient".

MML can only store 32 custom instruments at one time, which is presumably less than English's total base sound count. You'd need some kind of dynamic loader probably that keeps an in-memory bank of sounds and loads in which are neccesary to the sentence. This would of course limit the complexity of sentences: they could only use 32 base English sounds at one time (none of this is counting sounds that can be reasonably created with multiple sounds or MML effects). There is an up-side to this: instruments and microphone capture have a range of flexibility so you can manage vocal samples effectively.