6 ms·

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

120 points by xrd 5y ago

pizza 5y ago
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
- emicklei 5y ago
  The melrōse tool starts a local HTTP server which accepts the language to process. This is how the VSCode plugin communicates.
Rochus 5y ago
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
usrusr 5y ago
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.com/teg/102301/irc.html https://web.archive.org/web/20011204042933/http://www.spaz.c...
- skhm 5y ago
  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 https://github.com/harryaskham/funktor
- jrm4 5y ago
  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.
  - odyssey7 5y ago
    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 https://www.euterpea.com
    - jrm4 5y ago
      Nice. I defintely did not mean a "pure LISP," just something, you know -- parentheses-y or similar.
inciampati 5y ago
For the livecode-curious, the Algorave 10th birthday party stream https://ten.algorave.com/ https://ten.algorave.com/ is going on right now! https://live.eulerroom.com/ https://live.eulerroom.com/
anthony88 5y ago
I played with it a while ago and I was able to have the music to play on my synthesizer using a MIDI cable.
aaronats 5y ago
Very cool project! Look forward to playing around. Check out JSMdid if you get a chance. https://github.com/aaronats/jsmidi https://github.com/aaronats/jsmidi
bradrn 5y ago
Skimming the article, this appears very much like TidalCycles: http://tidalcycles.org/ http://tidalcycles.org/
cutler 5y ago
Overtone (https://github.com/overtone/overtone https://github.com/overtone/overtone), for me, is by far the most elegantly designed music software.
- bwanab 5y ago
  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.
  - cutler 5y ago
    Sadly true but Clojure doesn't change much so it still works.
ipnon 5y ago
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.
koide 5y ago
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?
politician 5y ago
Can it go the other way: input an audio file, output a Melrose approximation?
- pizza 5y ago
  Try a MIDI transcription neural network: https://paperswithcode.com/task/music-transcription https://paperswithcode.com/task/music-transcription
- emicklei 5y ago
  It is work in progress to capture MIDI input and produce a Melrōse sequence expression from it
eyelidlessness 5y ago
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!

120 points by xrd 5y ago

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

emicklei 5y ago
The melrōse tool starts a local HTTP server which accepts the language to process. This is how the VSCode plugin communicates.

Rochus 5y ago

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

usrusr 5y ago

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.com/teg/102301/irc.html https://web.archive.org/web/20011204042933/http://www.spaz.c...

skhm 5y ago
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 https://github.com/harryaskham/funktor
jrm4 5y ago
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.
- odyssey7 5y ago
  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 https://www.euterpea.com
  - jrm4 5y ago
    Nice. I defintely did not mean a "pure LISP," just something, you know -- parentheses-y or similar.

inciampati 5y ago

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

anthony88 5y ago

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

aaronats 5y ago

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

bradrn 5y ago

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

cutler 5y ago

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

bwanab 5y ago
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.
- cutler 5y ago
  Sadly true but Clojure doesn't change much so it still works.

ipnon 5y ago

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.

koide 5y ago

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?

politician 5y ago

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

pizza 5y ago
Try a MIDI transcription neural network: https://paperswithcode.com/task/music-transcription https://paperswithcode.com/task/music-transcription
emicklei 5y ago
It is work in progress to capture MIDI input and produce a Melrōse sequence expression from it

eyelidlessness 5y ago

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!