About MusicLang

The Python framework to write, analyze, transform and predict music.

What is MusicLang ?

MusicLang which simply stands for “music language” is a Python framework implementing a new language for tonal music. This language allows composers to load, write, transform and predict symbolic music in a simple, condensed and high level manner. MusicLang internally uses a `LLM (Large Language Model) <https://huggingface.co/floriangardin/musiclang`_ to predict what could happen next in a musical score. This framework is well suited to : - Generate musical ideas quickly. - Predict what could happen next in an existing midi file - Create an interpretable and information rich text representation of a midi file

Note

Writing music with this framework supposes that you have some basic knowledge on scales, tonalities and roman numeral notation of chords.

How to install

MusicLang is available on Pypi

pip install musiclang

A first example

1. Here is a simple example to write a C-major chord in musiclang and save it to midi

   from musiclang.library import *
   
   # Write A C major chord
   score = (I % I.M)(s0, s2, s4)
   
   # Store it to midi
   score.to_midi('c_major.mid')
   

2. Create, transform and harmonize a theme quickly

   from musiclang.library import *
   
   # Create a cool melody (the beginning of happy birthday, independant of any harmonic context)
   melody = s4.ed + s4.s + s5 + s4 + s0.o(1) + s6.h
   
   # Create a simple accompaniment with a cello and a oboe
   acc_melody = r + s0.o(-1).q * 3 + s0.o(-1).h
   accomp = {'cello__0': acc_melody, 'oboe__0': acc_melody.o(1)}
   
   # Play it in F-major
   score = (I % IV.M)(violin__0=melody, **accomp)
   
   # Repeat the score a second time in F-minor and forte
   score += (score % I.m).f
   
   # Just to create an anachrusis at the first bar
   score = (I % I.M)(violin__0=r.h) + score
   
   # Transform a bit the accompaniment by applying counterpoint rules automatically
   from musiclang.transform.library import create_counterpoint_on_score
   score = create_counterpoint_on_score(score, fixed_parts=['violin__0'])
   
   # Save it to musicxml
   score.to_midi('happy_birthday.musicxml', signature=(3, 4), title='Happy birthday !')
   
   # Et voilà !
   

3. Predict a score using a deep learning model trained on musiclang language

   from musiclang.library import *
   from musiclang import Score
   
   # Some random bar of chopin op18 Waltz
   score = ((V % III.b.M)(
       piano__0=s0 + s2.e.mp + s3.e.mp,
       piano__4=s0.e.o(-2).p + r.e + s0.ed.o(-1).mp + r.s,
       piano__5=r + s4.ed.o(-1).mp + r.s,
       piano__6=r + s6.ed.o(-1).mp + r.s)+
   (V['7'] % III.b.M)(
       piano__0=s2.ed.mp + r.s,
       piano__2=s4.ed.mp + r.s,
       piano__4=s6.ed.o(-1).mp + r.s,
       piano__5=s0.ed.o(-1).mp + r.s,
       piano__6=s4.ed.o(-1).mp + r.s))
   
   # Predict the next two chords of the score using huggingface musiclang model
   predicted_score = score.predict_score(n_chords=2, temperature=0.5)
   # Save it to midi
   predicted_score.to_midi('test.mid')
   

Please note that this feature is still experimental, it will only work with piano music for now and the model is not yet trained on a large corpus of music. If you want to help us train a better model, please contact us

4. Mix everything together to create a new pieces of music !

Learn MusicLang

To learn MusicLang we strongly advise to read the MusicLang’s user guide.

Contributing to MusicLang

MusicLang is a very recent library and is moving fast. Now it’s quite exciting times because the roadmap is still opened to change. Don’t hesitate to contact me. We are very interested to get in touch with composers, musicologists, programmers, data scientists and any other people who want to help us. We will regularly update issues on our github repository. Don’t hesitate to submit your own pull request if it makes sense for you and reflect your usage of musiclang.