Creating sheet music from existing MP3?

[exigence]

That's really what I'm asking. Can this be done?

[quantum]

It is most reliable when a human does it.  In this case it would be called transcription.  This is a good skill to practice.

[exigence]

It is most reliable when a human does it.  In this case it would be called transcription.  This is a good skill to practice.

Should have been more specific. Hand transcribing would certainly be best, but I was interested in a software approach first as something to get the bulk of the work done.

[ted]

I suspect the problem is still unsolved except for music which is very simple. There are several programs, e.g. Widi, whose authors claim can do this, and you can download restricted versions to try. I did so with Widi and tried it on my improvisation. Quite frankly the results were frightful. I expected rhythms to be distorted but most of the pitch recognition was also impossibly crude. I could do better by listening, and that's saying something because my ear's bloody terrible.

[exigence]

Yeah, you're right. I just tried WIDI -- no thank you. Was worth a shot, I guess.

[Bob]

I've seen software that will do a single line.  There was one I was looking at this past spring that would work with a single instrument's sound and could handle multiple notes -- It was $800.

[lostinidlewonder]

I looked into this when I was doing my Engineering degree and had to come up with a research project over 10 years ago (we also looked into scanning sheet music and creating a program to play the music). Still at the moment there doesn't exist any program that can do this accurately for us. Even midi programs which record exactly what you play on a digital piano will never produce notes in any program that resembles proper written sheet music (unless you play 100% exactly as written with an exact tempo playing synchronized with the recording program, which is simply an impossible feat). The intelligence that is needed for the program to be able to order notes in a proper format is beyond anything we can achieve at this moment. There is simply way too much error to deal with and ambiguous solutions, so no program will be really able to do it accurately in the end you will have to rearrange its estimated results into something sight readable.

[ted]

I am trying the demo version of this one on various recordings and it seems to recognise many more pitches than Widi. As others have pointed out, making a legible approximate score from the mess of pitches is entirely another problem, but I suppose if 90% of the pitches were right it would be a big help. If I were to use something like this, I would probably use the piano roll screen of the transcription (perhaps printed) in conjunction with the actual recording and do it in little sections at the piano. The score and the played transcription are far too messy to be of any use for a complicated piano solo, for example an improvisation. I  also found that the quality of the recording has a marked effect on the accuracy of the results.

Unlike Widi and the others though, this one does seem to have sufficient pitch recognition to be a help, admittedly with time, effort and the best recording possible.


https://www.sibelius.com/products/audioscore/ultimate.html

[Bob]

Audioscore is the product I had seen before.  I never bought it but I remember those eyes and Neuratron from several years ago. 

They might have a demo version available.

[Bob]

I am trying the demo version...

Duh on me. :p