.iesai Doesn't sound so difficult really. Why not having an algorithm translating IPA to lojban for all languages, so for every phone (phonetics) you translate it into the most lojbanic phonematic counterpart?
Lojban has a rather low number of phonemes, so there is no function mapping IPA to lojban. Also, the optimization problem for approximating sounds that Lojban doesn't have is in the general case ill-posed. Specific examples such as theta/thorn can be solved, but getting a general mapping from IPA to Lojban isn't going to happen. Nor is it going to happen with IPA to English, which has many more phonemes.
mi'e la latro'a mu'o