Well, I think you can’t theorically automate a conversion from phonetic to actual text because the correspondance between those two writings is not bijective:

For example if your phonetic word is [o], you may have to transcript it to 9 different forms:

eau, eaux, ô, au, aux, haut, hauts, oh, ho.

So you must use the intelligence of the text to transcript correctly from phonetic to actual text, and no application can actually do that without an error.