Jack is right: the subtle timing changes that performers add gets in the way. Anything like a midi to notation converter is likely to get a mass of hemidemisemiquaver rests, notes of equally odd timings and so on. If   the interface allows you to set minimum recognised lengths you can very easily fail to capture everything you need. I had an analogous problem splitting audio input into distinct tracks. The automated system detected periods of silence (or low noise) and if the silence as long enough it assumed a new track. This was a total failure on solo unaccompanied voices and I spent so much time merging tracks that should not have split and splitting those that should that in the end it was faster just to do the whole caboodle manually.