For the current or feature version of Dorico 5 PRO can you please ad some sort of AI feature that transcribes accurate full score notation from audio. For some of us with large projects this would save a lot of time. For example; If I exported an audio file of my full score I would like Dorico to read it and generate full orchestral notation for that piece of audio.
This is the equivalent of separating the eggs from the omelette. (It’s not going to happen)
8 Likes
Its a matter of oppinion @Janus - realize everyone has different workflow needs, especially those of needing to generate notation fast for media cues, broadcast cues and live theatrical pieces.
This has nothing to do with your workflow; it has to do with whether this can be accomplished.
I’m not saying it’s impossible, but at the moment I think it’s pretty close to it, for anything that’s more than trivial. Imagine just a string quartet: How would the program know whether a note is played by the first violin or the second violin? Or maybe by the viola? Now expand that to the whole orchestra. Is that one oboe, or two in unison? What if the clarinet doubles the oboe?
And even if this were feasible, it would be more likely to be handled by some other program that would save the result as MusicXML, just like scanning of sheet music is handled today.
3 Likes
No. It’s a fact. The audio stream contains sonic data from all the instruments and the ambience in which it was recorded that cannot be disaggregated.
2 Likes
Have you ever tried Demucs (which is used in Steinberg’s SpectraLayers) or any other source separation software?
It’s incredible what can be done already.
2 Likes
Software nowadays can do automatic recognition of rhythm instruments or voice, separate it from the rest, as well as Pitch Recognition to convert it to MIDI and Tempo recognition.
Now most of them don’t work 100% reliable, and the combination of all is tricky — but things can surely go this way quicker than one might think.
But worry not: I would not like the Dorico team to spend too much time on such features, which are much better handled by other departments within Steinberg or other companies.
5 Likes
Oh and if surely will not happen within the immediate to medium term future for sure.
Absolutely not coming at the very least in 10 years.
1 Like
It could be done now for a string quartet (probably almost perfectly for good recordings) with the right data for the machine learning.
Nooo!!! Breathe developer fairy , breathe!!!
2 Likes
No developers necessary for the separation process.
Demucs can sometimes already separate the instruments better than my ears can.
The Dorico development team is quite small and has a mountain of requests at their door, many of which are smaller potatoes than such a feature which would require a gargantuan amount of development resources, time, and money which doubtful Steinberg (parent company Yamaha) is willing to devote.
Many of us would rather development remain focused on Dorico being better at what it is designed for.
That said, there is hope in the future for third-party development, where I think something like this could be ideal for – a separate company with AI experience and resources, for example. So I’m not saying it’s absolutely impossible, but it’s not likely to be on Dorico’s radar as it would be better suited for a third party.
The implementation would require development, and the Dorico team has made it a point to never just slap features on.
3 Likes
I’m not suggesting that the Dorico developers should spend any time on this (other people are doing the work).
I am saying that anyone insisting that a computer couldn’t possibly separate the instruments in a recording of a string quartet is wrong. Obviously an orchestra would be much more of a challenge.
I look forward to you providing evidence (in the form of a notated score and original audio).
1 Like
I don’t know if any work’s being done on the transcription side but apropos source separation, you might want to experiment by running for example, some Lieder through Demucs.
It’s been trained on pop music but can sometimes make light work of separating a male voice and piano. Pianists selling accompaniments over the Internet are going to be out of work…
From my limited exposure, all manner of unwanted artefacts appear in the separated stems. I think accompanists are secure.
Suggest you google “audio to mid”. Many very smart people have been trying to get this to work - and for decades now - with no real solution. The first person/company that makes a working app to convert complex audio input to midi will make a lot of money.
1 Like
Melodyne?