Log in

View Full Version : How do you make an MIDI? (sorry, not sure where this goes)



MabiJaele
September 13th, 2011, 04:08 am
I'm trying to find or make a midi for this song: http://www.youtube.com/watch?v=sUChDL786vI
Corridors of Time (Chrono Resurrection) - Zohar & Mathew Valente Piano Remix

I don't know if it's possible to make a MIDI from an MP3. Is it?

I'm trying to make it usable on Synthesia since I can't figure out the sheet music even with the youtube tutorial (it has four clefs to read!!!!!!)

Could someone direct me to a website where you can make midis, give me instructions on how to make midis or somewhere where I can request a midi to be made?
Thanks!

M
September 13th, 2011, 04:20 am
MP3 -> MIDI = no-go. Software does exist that does this, and it's gotten much better over the years, but it produces a byproduct that is drastically inferior to the MP3 and typically incorrect sounding.
Sheet Music -> MIDI = go. While this takes time, you can take sheet music and plug it (that is, enter/engrave the sheet music into the software note-by-note) into programs such as Finale, Sibelius, or Musescore and they can ultimately export to MIDI.

On a personal note: I hate how difficult it is to work with MIDI.

There is one other option, too. You could request a MIDI/Sheet Music be generated off of that clip in the Sheet Music Request forum.

cryskolt_19
September 13th, 2011, 01:16 pm
Hey M, you're an expert in computer programming right? I just want to inquire with you, do you perhaps know why it is so difficult to design a program that allows one to transcribe MP3 into MIDI?

I'm no expert, but I do know that any note played on a musical instrument produces a sound with a unique frequency that can be picked up & translated numerically with a tuner.

I was thinking if it is possible to create a program that detects notes in a song by decoding the frequencies of the sounds with a tuner software. Then, after obtaining the notes, the program can construct the MIDI accordingly. Do you think that is feasible? :heh:

Emeraldshine
September 13th, 2011, 04:19 pm
There is technology that can "decode" frequency into a type of map (I think it's called frequency spectrum analysis), but the result is very messy, and you generally have to single out the pitches to be converted by hand. And say you have a note that starts very quietly and crescendos? How do you seperate that sound from background noise? What about overtones? What about a cymbal crash? How in the world do you convert that? Also keep in mind that computers are HORRIBLE at distinguishing between different types of instruments.

alpiso
September 13th, 2011, 08:56 pm
You can also try an OCR software. This kind of software take a sheet music and translate it into a midifile. So you just need to provide the picture (maybe a PDF, depending on the software) and you get a midifile.


I'm sorry, I don't remember the name of one of them, but I know they exist because I tried one many years ago :heh:

About automatic transcription, I don't remember where, but it seems there is already a discussion about this :think: ...

M
September 14th, 2011, 01:02 am
Hey M, you're an expert in computer programming right? I just want to inquire with you, do you perhaps know why it is so difficult to design a program that allows one to transcribe MP3 into MIDI?

I'm no expert, but I do know that any note played on a musical instrument produces a sound with a unique frequency that can be picked up & translated numerically with a tuner.

I was thinking if it is possible to create a program that detects notes in a song by decoding the frequencies of the sounds with a tuner software. Then, after obtaining the notes, the program can construct the MIDI accordingly. Do you think that is feasible? :heh:

I'll let you know once I'm done with my doctorate (this is actually one problem I'm trying to solve using genetic algorithms and machine learning). I've been seriously researching waves and harmonic series in terms of digitization to figure out how to exactly decode multiple pitches at once, and identify unique sounds within sounds.

The problem is pretty easy to grasp: A computer reproduces sound based off of a series of ones and zeros. No one sound can occur at the same time. The reproduction of that sound is created by using a decoder that understands these patterns and creates a square wave that modulates quickly enough to make it appear as smooth as the original wave. Similarly for listening to audio, a computer takes in data from a diaphragm (microphone) and creates a similar one and zero pattern as the decode phase produces, and records it as data.

So, if everyone plays the same note everything is good, as everything is in the same harmonic sequence. A computer would pick the sound up of everything and reproduce it similar to the way how it was performed.

Here's the real problem: what happens when two sounds occur at once?

When that happens, encoders and decoders quickly modulate between the two sounds as to make it appear like the notes are being played at once, when they are actually pulsating faster than your ears can detect. This speedy shift between tones causes translation engines to be confused if that is part of what's being produced, or if it's background noise, or if it's two notes being played quickly between one another. That confusion results in the garbled mess that classic audio to MIDI engines produce.



This is easy to demonstrate if you have a tuner by doing the following steps with two of the same instrument:
Play the same note on both instrument such that the tuner picks up both players.
Have the first player go up a half step. Notice the tuner moves to the sharp side of the note (about 40 cent).
Have the second player go down a half step. Now notice the tuner is bouncing between sharp and flat

This movement is an example of how a computer is unable to determine who is exactly playing what.

alpiso
September 14th, 2011, 06:21 am
this is actually one problem I'm trying to solve using genetic algorithms and machine learning

Wow O.O ! Great :P

I did some studies on genetics also, this point of view is interesting ! I was stopped with an algorithm base on the fourrier transform (http://en.wikipedia.org/wiki/Fourier_transform), as I studied it too.

I found the thread (http://forums.ichigos.com/showthread.php?17628-mp3-to-midi) I was talking about ;)

cryskolt_19
September 14th, 2011, 07:09 am
Thanks everyone. All these makes the human ear seem like a miracle invention. :lol: