There are two completely different representations of music that are used widely in the music industry:
Audio representation of music contains exactly the sound that was originally recorded or produced synthetically. It does not contain any additional information - no score, words or even the list of instruments that were played can be obtained directly. Modification of Audio are mostly limited to applying general effects like equalizer or reverberation. All audio events are mixed to one audio stream and there is no way to reconstruct or modify a particular part only.
MIDI representation is somewhat close to the classic musical notation. MIDI track consists of commands that indicate which notes should be played with which instruments and when. MIDI track can be played with a MIDI synthesizer (either software or hardware), and the sound greatly depends on the actual synthesizer used. MIDI tracks cannot contain vocal parts with words or similar sounds that cannot be represented via MIDI commands.
MIDI to Audio conversion canbe accomplished by simply playing MIDI track and simultaneously recording it as an Audio track. There are also third-party offline MIDI renderers than can do the process more efficiently.
Audio to MIDI conversion, on the opposite, is a very difficult task of guessing what notes where played in a recorded piece, and this is what WIDI Audio To MIDI Audio Unit is designed for. WIDI AU uses complex mathematical algorithms that identify sounding pitches and detect their starting and ending times. Additional techniques are used to recognize chords, cut out overtones and improve MIDI quality. Because of the process nature, compositions are often not recognized perfectly, but MIDI data can be edited easily and conversion errors can be corrected manually.