My boss has asked me if there’s a bit of software that will take a piece of music as audio and transcribe it as sheet music.
It seems to me that even for a relatively simple piece of music, it would be a tall order for a computer program to make a good job of this and for something complex like an orchestral piece, or music and vocals with harmonies, I can’t imagine it being done on anything less than a supercomputer.
Still, maybe someone out there knows different, so I thought I’d ask.
The higher-end notation software packages like Sibelius can transcribe music from MIDI-keyboard input while the user plays. There might be software to convert MP3 or other digital audio to MIDI format. But you’re right, it is a tall order to go straight from audio to instant sheet music, although it would depend on the music’s complexity.
I’ve used the keyboard-to-notation midi interface before, and I have to say it’s a nightmare. I’d imagine that any audio-to-notation would be bad too.
The reason is that when a musician plays written music, they add their own nuances of timing into the piece. The computer, being a dumb box, will record these variations absolutely.
Thus if I linger on a note just a few fractions of a second beyond the end of a bar, the computer will add a few slurred hemi-demi-semiquavers of the note into the next bar. This is fine for a computer to read and reproduce, but near-impossible for a human to interpret.
There may be software out there nowadays that guesses what the musician was actually reading, but I haven’t heard of any. Also, if more than one instrument were involved, IMO there’d be an exponential amount of complexity for the software to handle. Someone correct me if I’m wrong.
I’ve used those too, and the trick to getting decent results is to use the “quantize” feature, or whatever it might be called on your particular software. In effect, you tell the software to “round” to a certain minimum note length. If you’re playing a piece that is not overly complicated, you can get good results rounding to the nearest 8th or 16th note - that will take care of many of the problems you describe. Still won’t be perfect, but often it’ll be good enough that you can fix the remaining mistakes by hand.
I have to say I’m skeptical about Intelliscore’s claims. I’ve seen software that had mighty impressive claims for doing this before and they generally produced a mess for all but the simplest of music.
Even if it did work; you’ve got a MIDI file, not sheet music. As jimm explains, a midi file converted to a score does not necessarily mean a score that’s in any way sensible to a human reader.
Still, if it’s good it’s probably faster than doing it yourself from scratch, unless you’re one of those annoying musical geniuses who can transcribe directly by ear at will. I might try it out.
Actually, I’ve tried IntelliScore. Bad. Bad. BAD. I downloaded their demo and tried to convert a simple piece of monophonic video game music to MIDI just to test it. The software required you to tap your spacebar in time with the music, which I found quite absurd. When the MIDI piece was complete, I played it. To say in the least, the music wasn’t very much fun to listen to. There was nothing in the MIDI that resembled the original MP3.
Honestly, MP3-to-score transcription via computer simply isn’t possible with current technology. We’ve barely got voice recognition down.
As I think about it, it doesn’t seem that there would be any real obstacle to making this happen. Except, perhaps, for a lot of money and a lot of time.
It would be the equivalent of having voice recognition software that could run on a computer at a crowded party and track 8 simultaneous conversations of 2-5 people apiece, in more than one language, and separate them out into distinct sentences uttered by Person 1, Person 2, etc.
I would not say “It ain’t gonna happen”. I suspect it is gonna happen, in fact. But I wouldn’t count on a functional version before Worldwide Developer’s Conference this year or anything.
Plenty of real obstacles. Simply adding more time to crunch the waves isn’t going to change that.
Want it to pick out the flute solo in the middle and put it into a seperate track. Great. What’s a flute sound like? OK, you give the computer a sample of what a flute sounds like. Except this flute has reverb, and a guitar following the same melody. OK, give it a guitar sample and explain reverb. Hang on, wrong guitar sound, it’s more like this…
Pretty soon you personally have had to break the original track down into small enough sections for the computer to understand that you would have been as well doing the whole thing yourself. And the on next track you have to start all over from scratch…
It’ll be far easier (and probably cheaper) to find a human who will do the work by the hour. Sure, computers would be capable of doing this, but nobody has yet writted suitable programs.
Thanks guys. I researched this for my father-in-law a few years ago, and found nothing, and it is a relief to hear that IntelliScore doesn’t do it either. He is a composer (still going at 86!) and uses the MIDI keyboard to Sibelius interface. It works well for him, but he is doing a line at a time, not really playing a long piece into it.
Sibelius has probably added years to his life. He’d never be able to write music by hand anymore. The only downside is that when he comes to visit he monopolizes my computer, working on his latest piece.