notating music

[brokenagraffe]

How does someone notate music from music (audio) files on their computer? Is it accurate? I'm trying to get a certain transcription for solo piano of a Tchaikovsky song (Op.47 No.6) but it's nowhere to be found... so I figured computer notation could work even though I've only heard of it being done.  I have no clue how it works so I need a thorough explanation if possible. Mucho thanko in advance hombres.

Note: I have shitty ears so I can't just listen and go to the keyboard and play it, ya know?

[pianonut]

if this is possible, i'm interested, too. i want to transfer to sheet music some cadenzas! 

[ted]

The last time I enquired about this it was still an unsolved computer problem, possibly unsolvable. Many people confuse it with the graphically complex but easily solvable problem of producing notation from a file of discrete notes (midi, digital file etc). It seems what we want to do is find what large set of component waveforms ,added together at variable times and volumes, is likely to have produced a resulting complex waveform. A couple of tries have convinced me that if this analysis is possible at all it is far from simple. You can write something which will guess at a couple of notes, but a complete complex solo ? Not by a long way.

Can somebody knowledgeable here comment further on this question ? Is it  something which people are working on ? Is the waveform the only possible computer representation of sound ? I too would very much like to know some answers. The question intrigues me.

Of course common sense tells me that people like Dapogny, Posnak, Scivales and others wouldn't have wasted months picking up complex solos by ear if an electronic means had been available.

[xvimbi]

Well, I don't think it is quite possible. One can easily deconvolute sound into its components by means of Fourier analysis. This way, one gets a collection of all the waveforms in the sound in form of their frequencies and amplitudes (essentially loudness). One could probably bias that Fourier analysis such that the components will only be frequencies that correspond to the notes on a piano, plus their overtones. This is where I see the problem. The sound that a piano makes is not pure; there are all kinds of frequencies present that do not correspond to regular pitches. I don't know. Somebody should sit down and write such a program. I bet it's been done already, but I don't know how successful something like this is.

[ted]

There's an interesting article about it at the following link. I begin to have a new respect for the human brain and ear !


https://ismir2002.ismir.net/proceedings/02-FP01-2.pdf

[ted]

Okay, this has got me thinking now. If waveform analysis is so hideously impractical, why couldn't it be done by having an electronic instrument comprising a battery of eighty-eight sensors, each similar to an electronic tuning device ? Each device could be made sensitive to one piano pitch and the responses while a piece was being played fed into a computer. There must be something wrong with this idea because it's so simple somebody must have thought of it.

[Derek]

I tried opening a midi file in Finale the other day. One would expect this to translate perfectly into Finale's notation, since Finale itself can produce midi from notation.

However a lot of bars and things spilling over into other measures occurred, even though this was a professional musician playing a baroque piece at an even tempo! He clearly wasn't playing EXACTLY on the beat.

Anyway its becoming clear to me that raw MIDI data seems to have a perfectly accurate representation of what is played by a human being, down to the tiniest imaginable subdivisions of time (probably tinier than we can be aware of),   but this is very different from a computer being able to figure out exactly when a measure begins and when one ends. It seems to me that one could tell a computer "count this many notes and fill each measure with them"  at least for certain kinds of music, like the baroque piece I mentioned. 

[ted]

"the graphically complex but easily solvable problem of producing notation from a file of discrete notes (midi, digital file etc)"

I was a bit hasty making that statement, Derek. I meant that it gets the pitches plus their distance apart in time in "some sort" of notation because the pitches are already "there" in discrete form. Rhythmically meaningful conventional notation ? No, of course not. I once improvised on a keyboard and had the results turned into notation. Everything except the pitches looked a ghastly mess - completely impossible for somebody to read meaningfully.

I had assumed the original poster was thinking mainly of getting all the pitches played in a complex solo. The position with regard to rhythm would be even more intractable because notation cannot capture anything but the simplest felt rhythms anyway.