Melrōse, a language to program and play music (2021) | Hacker News Reader

Melrōse, a language to program and play music (2021)

ernestmicklei.com

120 pointsxrd21 comments

Nice, another entry to the tiny list of musical computer languages that stick to notation level and don't go down the rabbit hole of synthesis only to never return.

  notemap('!...!...!...!', kick)

Can't help thinking of this classic: https://web.archive.org/web/20011204042933/http://www.spaz.c...

i have a haskell DSL [1] (really all the heavy lifting is done underneath by csound-expression though) where this looks like e.g.:

  -- 4 on the floor bass drum, on for 16 beats, off for 16 beats
  kicks <- do
    d <- drums "X _ _ _|O _ _ _|O _ _ _|O _ _ _" Tr808.bd2
    env <- sqrEnvM 0 16
    return (d & withEnv env)

nice to see the pattern again in the wild.

[1] https://github.com/harryaskham/funktor

Oh wow

Yes, this.

I'm only a casual music maker and I could never quite put into words this "problem." I just always wanted a functional (in the LISP sense, not just "usable"), flexible way to trigger notes (and by extension, samples and loops), and you're right, pretty much all of them veer into synthesis, which I don't much care about AT ALL. We got tons of samples and ways to make sounds without twiddling envelopes.

Cannot wait to get some time to find some of these and play around.

Though Haskell isn't a lisp, the book Haskell School of Music teaches it by exploring the authors' Euterpea music library.

https://www.euterpea.com

Nice. I defintely did not mean a "pure LISP," just something, you know -- parentheses-y or similar.

Interesting; a discussion would help about the core benefits compared e.g. to Sorensen's work like Impromptu, or ChucK, or OpenMusic.

> Melrōse offers a fully human-centric approach to algorithmic composition

Rather a "programmer centric" approach

For the livecode-curious, the Algorave 10th birthday party stream https://ten.algorave.com/ is going on right now! https://live.eulerroom.com/

You could imagine some GitHub Copilot like model that is trained on available music then uses a language like Melrose to predict music. That way the model sends notation to a synthesizer instead of directly creating a waveform. It would prevent the odd artifacts currently present in GAN music.

This looks really cool! I’m particularly interested in the event-driven APIs: for years I imagined writing partially live/partially pre-programmed music where the latter adapts to the former by aspects of its performance alone, to give even the pre-programmed music more of an improvised feel. Eg key and tempo changes, extending a bridge/solo. This looks like it can achieve all of that, given some creative application of MIDI signals. I’m excited to try it out!

This is cool - just wondering, it's written in go; what's the best way to interface it w/ Python? In order to e.g. combine it w/ a ML model in say pytorch

The melrōse tool starts a local HTTP server which accepts the language to process. This is how the VSCode plugin communicates.

Skimming the article, this appears very much like TidalCycles: http://tidalcycles.org/

I played with it a while ago and I was able to have the music to play on my synthesizer using a MIDI cable.

It surely is my limitation, but I have a hard time of visualizing uses for these tools other than interfacing them to ML models (which is not a very compelling use case in my opinion). What other cool things have been done with programmatic music tools?

Very cool project! Look forward to playing around. Check out JSMdid if you get a chance. https://github.com/aaronats/jsmidi

Overtone (https://github.com/overtone/overtone), for me, is by far the most elegantly designed music software.

Unfortunately, it is pretty much a dead project. The creator determined that clojure was the wrong language for a wide audience and has sense created Sonic Pi which is programmed in Ruby.

Sadly true but Clojure doesn't change much so it still works.

Can it go the other way: input an audio file, output a Melrose approximation?

Try a MIDI transcription neural network: https://paperswithcode.com/task/music-transcription

It is work in progress to capture MIDI input and produce a Melrōse sequence expression from it

Reply on news.ycombinator.com