Let me say that I am simply gobsmacked that we have MP3 to MIDI conversion at all.
The two are completely different ways of recording music. MIDI is akin to a sheet music score. WAV is a digitization of the actualy audio signal. MP3 is simply a lossy compression for WAV (“lossy” means that the input signal before compression is not identical to the output on playback; however a great deal of thought went into making sure that the parts that are discarded make little difference to the human auditory experience)
A skilled musician may be able to listen to music and jot down a reasonable score, but it’s a pretty impressive act. As impressed as I am by the fact that we can do it at all, I wouldn’t expect too much from the results. I presume they are mostly of use to MIDI users who can ‘clean up’ or modify the resulting MIDI score. I used to work with MIDI as an amateur, from the 70s to the end of the 90s, and good reliable audio-to-midi conversion has been a Holy Grail since day one.
Not all music CAN be converted directly to MIDI in any reasonable fashion. First, ithe program must assign the instruments to score, and there might not be any available instrument that matches the studio-modified music, or a method that might occur to a skilled human might not occur to a machine. Human voices are probably the best and commonest example of something that can’t be MIDI’d well.
Think if it this way: a program doesn’t “know” what it’s listening to. You could feed it the sounds of a construction site, and it would generate an output – but I ask you, is a reasonable music score of that sound possible? Even with hundreds of standard musical instruments and skilled composers, could anyone write a symphony that accurately reproduced the sounds of a construction site.
Another problem comes from timing: real human-played music doesn’t hew to the precise clockwork metronome of a musical score. I’ve made many a MIDI by having musicians play MIDI instruments directly into a computer – and the resulting score sounded great – but the resulting score was a complex tangle that no musician could every hope to play back accurately. A human written score is a deliberate approximation; but a human-played score is immensely more complex: think of how easily and perfectly computers have long handled text, and how difficult it has been to get them to do mass-scale OCR. Irrelevant details screw the process up.
While some music will be simple enough for a good workable translationIt’s quite likely that the resulting MP3 will be far larger and complex than necessary, as the software tries to make a good approximation by whatever means possible (e.g. by using 100 instruments, where there might originally have been only 12). A human could clean it up fairly easily, but your cell phone may not be able to hold it, and your cell phone’s limited CPU may not be able to handle it.
I wish you luck in you endeavor – and I hope that audio-to-MIDI software has made enormous breakthroughs since I worked with it. But you might get better results faster by trying to find a MIDI, made and released by an amateur, than trying to convert your own through software. (I’m not sure of the various legalities of releasing MIDIs - whether thranscribed by a human or machine - but it’s pretty ommon on the Web, especially for popular music that’s been around a while