Audio file to MIDI conversion

George Grella's icon

I've been looking for software that will take an audio file, like mp3, and convert it to MIDI or even give me the notes in notation, as if I had transcribed it by ear. So far, I've turned up nothing for Mac OS X other than tools for standard transcription [yeah, I can do that, but some things are complex and I'd love to save time]. Has anyone done this, with or without Max? Any ideas or advice?

TIA

Mattijs's icon

My first reaction is that if you make something that does this accurately, you'll be a rich man.

For an mp3 of a solo piano played one note at a time, it might be doable though. Don't know of any program that does this automatically (but never really searched for it either).

Mattijs

Quote: George Grella wrote on Wed, 02 May 2007 23:06
----------------------------------------------------
> I've been looking for software that will take an audio file, like mp3, and convert it to MIDI or even give me the notes in notation, as if I had transcribed it by ear. So far, I've turned up nothing for Mac OS X other than tools for standard transcription [yeah, I can do that, but some things are complex and I'd love to save time]. Has anyone done this, with or without Max? Any ideas or advice?
>
> TIA
----------------------------------------------------

wippen's icon

A few (3? 4?) years ago, I found a couple of demos which claimed to do this, and I tried them briefly on an ensemble piece and a solo piano piece. (they were for PC). They spit out alot of notes which seemed vaguely related to the audio material, but I didn't take the time to examine them. I imagine that they could serve as a basis for some sort of transcription, but still haven't been able to compete with the human brain.

In agreement with Mattijs: if anyone could do this, they would definitly be able to buy a better piano than the one I have...

Stefan Tiedje's icon
George Grella's icon

Thanks for the responses. My unspoken thought was that this is something that still needs tackling on a software basis.

As a musician, I've been transcribing for over 20 years, so I understand the complexity and the decisions the ear and the mind make. I think it's possible to work this out on a software basis, especially with commercially available music tracks. Quite a project though! Maybe I'll get to it someday . . .

This thought all started for me about a year ago when I was trying out a transcription application, Transcribe!, I think, to do the task for a Cecil Taylor piano solo. I remember the application being a complete disappointment, really terrible. There has to at least be something good that handles the basic tasks of playing back specifically for transcription; isolate segments, repeat, variable-speed playback without altering pitch, etc.

wippen's icon

Quote: George Grella wrote on Thu, 03 May 2007 10:26
----------------------------------------------------
>There has to at least be something good that handles the basic tasks of playing back specifically for transcription; isolate segments, repeat, variable-speed playback without altering pitch, etc.
----------------------------------------------------

I haven't had to do much transcribing since I started with Max. When I was, however, I used to listen to things using Wavelab, primarily for the graphic simplicity of the playback. Your suggestion is a good one; it would be easy to make a good transcription tool using Max, although it would not be significantly different than the possibilities offered by Sibelius, Finale, Cubase, Logic, etc. The only elements missing in those programs (as far as I know) are variable-rate playback and configurable tone and chord recognition. Max, on the other hand, doesn't offer a score gui. It might be worth the trouble to have playback and certain recognition features controlled by max but send the information to another program with a strong graphical application.

Stefan Tiedje's icon

George Grella schrieb:
> As a musician, I've been transcribing for over 20 years, so I
> understand the complexity and the decisions the ear and the mind
> make.

I bet you will be more effective than any software for the next 15 years
at least...

> There has to at least be something good that handles the basic tasks
> of playing back specifically for transcription; isolate segments,
> repeat, variable-speed playback without altering pitch, etc.

If you do it non-realtime, you have all the time of the world. To get a
piano track transcribed might be possible, I'd actually try to play
along to check if the analysis is correct...
But thats a single instrument of known color. Even if you would assume
that all pop music is just using quicktime musical instruments as sound
source, you won't be easily able to distinguish an oboe from a string
sample within a mix of other sounds. This is even impossible for most
humans, you need a lot of listening experience to be able to do that...

I guess it needs a complete brain size neural network to get this going,
and that neural net probably needs the same 20 years to be trained as
you did... ;-)

We are quite close to be able to imagine such a beast, but its still far
away...

Stefan

--
Stefan Tiedje------------x-------
--_____-----------|--------------
--(_|_ ----|-----|-----()-------
-- _|_)----|-----()--------------
----------()--------www.ccmix.com

George Grella's icon

My approach to all this is as a jazz musician - I used to make transcriptions of instrumental solos for practice, now I'd like transcriptions of more complex music to have some raw materials for compositions. It's not so much the Finale idea, playing an instrument along with a click-track and a pre-defined meter and letting the software write the MIDI notes, but just to get the pitches themselves, and then let the hand and brain sort those out. Jazz players used to do this with variable-speed tape recorders, which of course changed the pitch, but with digital media it seems to me half the problem has been solved. Now, onto the other half!