Sigh. Have you ever tried to play a MIDI guitar? You have to be absolutely spot on with your playing, or the device sends the most weird MIDI info. MIDI is a VERY SPECIFIC protocol. All it does is sends a series of data packets. It needs the tempo to be correct first - although you can just whack the tempo right up and send a stream of data. How the device the other end responds can be difficult to predict, but a typical message will say Bb - played at XXX time, with YYY velocity and it stopped at ZZZ. A bend is a totally separate thing, sent as a stream of data with individual settings, over and above off sent sequentially. Aftertouch, which is a way of increasing and decrease the initial note on value is another stream of data. You can also send controllers that can control almost anything, IF, you tell the devices at both ends what they are. MIDI only has 16 separate channels of information in the entire stream. It's possible to also lose data - sending wild amounts of pitchblende and aftertouch, for example clog it up and if you then lose a note on, or worse, a note off message, it can get manic. Once you send the data, you need something at the other end to convert it back to audio. Currently, there is no device that can be sent say an mp3 of a band playing, and then extract bass, guitar, drums and keys. Too difficult. Is that a guitar playing an E, or the keyboard? No idea. The best ones at the moment are as I said, ones designed for guitars with a maximum of 6 notes at a time. MIDI is also tricky when it comes to timing. If the guitar sound starts a fraction late, it rarely matters, but if it's the snare or kick drum, it's pretty vital that it plays at the right time, so MIDI plays tricks and processes channel 10, where drums live, first - to give them a head start.
MIDI is the language of the digital stream of data. It is NOT, and never has been, a system to capture the entire audio spectrum. It was designed so you could connect a Yamaha keyboard to a Roland one, and when you played a C on the Yamaha, the Roland played the same. You could then record it, process the data and play it back. That is what MIDI still does. You cannot send a voice through MIDI systems. You could identify the note, then send that note's start and stop down the cable and have a synth play a string sound. As for anything else a voice can do, forget it. It's not like comparing a .wav with a .mp3. MIDI is nothing like that at all.
The Ableton clip, for example detects monophonic lines, so you could take say a bass guitar riff and produce a bass synth - but it is not converting sound into anything other than notes. The results are usually terrible.