A python library for generating random music to be used as training set for instrument detection and score-generation from recorded audio with machine learning. As presented (recording) at EuroPython 2018.

Truth is ever to be found in simplicity, and not in the multiplicity and confusion of things.

Isaac Newton

The benefits of using a music generator as training set over recorded music are:

• We can control the complexity of our problem (how many instruments, what type of instruments, type of rhythms, polyphone vs. monophone, etc.).
• We always know the ground truth.
• Since we are generating the data, the data set is basically infinite.

The package offers:

• Basic music theory:
  • Notes and their frequencies (tuning of A4 other than 440 Hz is possible, only the scale-invariant, equal temperament is supported for now)
  • Scales: major/minor scales, a generic scale is also available
  • Chords: major/minor chords, augmented and diminished
  • Tempo, time signatures, measures
  • Tracks and scores
• Synthesizer module:
  • Basic oscillators: square, sine, additive sine (to be added: triangle, wavetable synthesizer)
  • ADSR (attack-decay-sustain-release) enveloping for any generator
  • Butterworth filters
• Random music generation module:
  • Generate random melody using random walk

Create virtual env:

$ python3 -m virtualenv venv

Activate it:

$ source venv/bin/activate

Install the package:

$ pip install -e .

Run tests:

$ py.test tests

For installing simpleaudio you might need to install the python3-dev package.

On Ubuntu I also had to install:

sudo apt-get install libasound2-dev
sudo apt-get install python3-tk

You might need to install ffmpeg.

Generate music some music:

$ python scripts/gen_random.py