Vision: AI Assistant for Harmonic Analysis and Counterpoint in Dorico
Here’s a draft vision for an AI-based assistant integrated into Dorico that goes beyond traditional proofreading.
Not just checking notation rules, but interpreting the score musically — harmony, counterpoint, voice-leading, register, and spacing.
A tool designed especially for final checks before publication, with awareness of:
harmony and harmonic continuity
counterpoint and voice-leading
register, spacing, and balance
visual clarity and how musical structure is communicated on the page
The goal isn’t to replace the composer —
but to act as a musical proofreader and second pair of eyes inside Dorico.
This is a vision for implementation at a later stage, but it may already be useful when considering the direction of future development for Dorico’s proofreading tools.
I think that’s a very good idea. I’m an amateur composer/arranger, and recently I hired a teacher who reviewed my work and gave me lots of tips and suggestions. Unfortunately, that course is now over, and I really miss those suggestions. Recently, I uploaded one of my arrangements as a PDF to Google Gemini and asked what I could improve. From my amateur perspective, I got some pretty good points out of this. How much more could a specialized AI in Dorico deliver!
there was a massive thread about checking things like parallel fifths and octaves a while back. I was among many who don’t myself want such a tool. We all have different ideas about to what degree formal analysis can help in composing (in my case not at all) but having said that, my main concern would be it could slow Dorico right down.
I don’t really see this as an extension of proof-reading, though. Proof reading is designed to improve the readability of the score and most of its suggestions are eminently sensible. Specific practical things like unplayable notes are also fine. But it’s a different thing trying to apply analytical theory which many composers pay no attention to nowadays anyway. Still, there may be grey areas here and I wouldn’t want to get in the way of something which had widespread support providing it doesn’t take development time away from the many things still to be sorted out and. doesn’t negatively impact on system performance.
For one thing, imaging an AI proofreading assistant that understands your choices. Like when you discard one proofreading flag, it would askyou if you want to discard all similar flags, or look closer at corner-cases.
my understanding is that the previous comment was only to remember your choices about the kind of proofreading you clearly don’t want. We’re not discussing analytical tools here – indeed one of my own worries about this sort of thing is indeed that it might encourage the sort of formulaic composing I see far too much of in commercial music.
I’ll bet there’s plenty of Doricians who could use such a tool – not the professional composers and engravers, but students and less experienced users. Each time an amateur composer has something flagged, hopefully they’ll learn something to avoid the hassle of “fixing mistakes”, and choose whether to implement it.
This would also make it attractive for students to use Dorico. The more users, the better for all of us. Hopefully it would work in the free version to attract new people.
Maybe such functionalities would be useful for beginners who are not fluent with music theory indeed. In any case, the best proofreader is to listen to the audio rendering. The most important is that it sounds great according to my personal taste. I took years of harmony and counterpoint theory lessons, so I am quite proficient in these matters, but to my view theory never helped me to write music. I think theory is there to understand why some chords progression, voice leading or countermelodies do not sound great and theory helps you to identify which voice and notes create the issue and how to fix that easily. So functions that can highlights issues could indeed help beginners to quickly fix some issues.
I don’t know your background so forgive me if I’m stating things that are obvious to you, but there is so much misinformation in the AI hype bubble that I think this needs to be said:
A tool that you describe is certainly possible* (I’ll stay out of the discussion about whether it’s useful or not), done through traditional programming methods. It would not be doable through AI, as the term is commonly used today (large language models).
*When I say that it’s possible, I do so with the assumption that your blog post is using marketing speech. “It doesn’t merely check rules – it interprets the score musically.” does not make sense.
Edit: Oh, I read the “about” text on your blog. You are talking about having an LLM doing this. “Reflections through conversations with ChatGPT”. That’s not how the tech works. ChatGPT is lying to you.
It could be something for a teacher that has to correct student exercises/homework. But for a composer this has little value: a lot of music tries to circumvent these rules.
Young Domenico Scarlatti could have done with such a tool. It would have helped him avoid mistakes like this (K.301), and then he might have become a famous composer!
I’m aware that many people in the notation field are sceptical of AI in general — and of AI as a tool for music notation in particular. That’s fair, and it’s something I try to keep in mind.
That’s also why I think any AI-based functionality should be optional rather than core, at least for now.
I do believe this is technically doable. Whether it should be done — and by whom — is a different question.
For my part, I’m simply sharing a vision of what might be possible, and leaving it at that.
And not just in the notation field. In academia, it regularly hallucinates references. In “Vibe coding”, it often produces code that is erroneous or that introduces needless complications.
So, massive amounts of development effort into something that’s optional?
Anything is possible if you throw enough money at it. But who’s going to spend it, and for what return?
I, for one, would like an AI that could read crappy scans of sheet music with 99.99999% accuracy. But, as they say in Scotland: if wishes were horses, beggars would ride.
Look at eg. Affinity Suite integration with Canva.
The optional thing I feel is mainly to cater for those who are a little fanatic about not using AI
Anything is possible if you throw enough money at it. But who’s going to spend it, and for what return?
Well, in other fields there are no shortage of companies wanting to spend a lot of money on AI development. And for something like Adobe and Affinity, not doing so, would probably mean they would be out of business fast.
As a retired college music teacher who still teaches online, I find that AI has helped to create students with no problem-solving skills. With AI, there’s no incentive- or need- to develop any. To Ben’s point, even when students are told not to Google, use Chat GPT or AI for answers in the online course I’m currently teaching, they do. And, their answers are almost always wrong - sometimes hilariously so-because there is no contextual relationship between AI and the course content. The students seem to have no idea that this is so. Using the assigned course readings and watching the videos to look for the correct answers is apparently beyond their ability- or desire. My biggest concern about AI as it gets bettter and better is just that- people will become too dependent on it for too much and lose the ability or desire to find solutions on their own without it. AI can very easily become not just a tool, but a crutch.
It certainly has. It depends on a person’s sense of integrity and self to a degree I suppose. Always has - and will. Although, back in the day (I just turned 71 yesterday), if we had any hope of doing well we had to go to the library, use books, tapes, encyclopedias, periodicals, etc. to find solutions. We had to learn how to look for and find answers on our own. Of course, even then, some students didn’t do that and suffered the consequences.
I have no problem using AI as ONE of my tools for researching something, but it will always be just that for me -a tool in my quest for solutions to something -not the only thing I look at. It seems many students now don’t know how (or care) to look for answers outside of AI anymore.
We’re all supposed to see the big AI Assist when MuseScore 5 rolls out—if that ever happens. I’m guessing that the reason it’s two years late might be that AI assisted notation is not quite working out for them despite the million$ they are throwing at it.
I was hoping that Muse Group would show up at NAMM this year but, except for Hal Leonard, they were nowhere to be found.
