davidkant / mai

Music and Artificial Intelligence

Music and Artificial Intelligence

This code repository is the home of the Music AI Tutorial, a collection of cloud-based, interactive tutorials that teach topics in music, artificial intelligence, and computer science. The tutorial takes a historical approach. My intention is to give cultural, technical, and aesthetic context to contemporary approaches to Music AI.

Viewing the tutorials

The tutorials are intended to be viewed using Google Colaboratory, a cloud-based Jupyter Notebook environment, which allows you to edit and run code in the browser. You can also view a static render of the tutorials via nbviewer, or an interactive versions via MyBinder.

Teaching

The Music AI Tutorial was developed for the course MUSC 80L: Artificial Intelligence and Music at UC Santa Cruz Course with instructor David Kant. I am actively seeking new opportunities to teach this course. Please contact me if you are interested!

Thanks!

This project is in ongoing development, so please check back for updates!

mai's Issues

markov chain lecture sequence

find a way to get the lecture sequence into the notebook 4.4

microtonal MIDI

add support for microtonal MIDI (floating point pitch values) in music_makers

Python 3 please

music notation

using trnzscrybe + abjad

ImportError: cannot import name 'SGD' from 'keras.optimizers'

To solve this issue, go to nn.py and change from keras.optimizers import SGD to from keras.optimizers import gradient_descent_v2. And then in line 20 and 45, change SGD(...) to gradient_descent_v2.SGD(...).

which setup needs to be done?

install external libraries for sound playback

!pip install -q pretty_midi
!pip install -q pyfluidsynth
!apt-get -qq update
!apt-get -qq install -y libfluidsynth1

consider a setup function...

add curves to synth ADSR

replace np.linspace with scipy interpolate for curves

add MP3 rendering useful for longer files

Issue installing mai locally

I was playing around with this library recently using jupyter notebook locally through vscode and encountered this issue where mai.synths and the other functions were not properly installed after pip install mai . To fix this issue, I had to download a copy of this repo and drag and drop all the .py files to the locally installed directory. Ex: C:\Users\Austin\AppData\Local\Programs\Python\Python37\Lib\site-packages\mai\

This issue doesn't seem to occur on google colab.

microtones don't reset on `int`

In pitches = [60, 60.5, 60] the last note 60 will sound as 60.5 b/c pitch bend is not adjusted on microtones=0 which is not OK when mixing ints and floats in a list of pitches.

make_music_heterophonic PLOT

plot function for make_music_heterophonic

multiple instruments

music_maker with multiple instruments

realtime MIDI rendering

a pipedream

write a custom essentia feature extractor

write a custom extractor so we dont extract redundant features AND you can start from 0 not this off by frameSize/2 business.

make music notation faster

Removed music notation support because linux setup is too slow. Specifically, the two requirements below take about a minute.

!apt-get -qq install -y libfluidsynth1
!apt-get -qq install -y lilypond

And then don't forget to
%load_ext abjadext.ipython
but that's not a bottleneck...

essentia frame_to_sample

essentia framecutter starts from -frameSize/2 which makes conversions from frames to sample tricky, especially when rearranging audio (like by ascending spectral centroid) and re-analyzing.

dependencies

add them