ScaileAI

A Python project used for creative purposes. The application generates chord progressions in the MIDI format through a LSTM recurrent neural network. The application is setup as a local server structure waiting for HTTP request. It works in conjunction with the Scaile app but can be expanded to a website environment.

• Developed in Python

Purpose of the project

• The purpose of the project was to explore if it was possible to generate chord progressions from a Neural Network. The object was to maintain the harmonic elements of the training data and generate a variety of chords with a sense of knowledge in music theory. Essentially matching the harmonic structure of a chord progression.

Generating

• By running the main.py project the training will start. This requires a path to a directory loaded with MIDI files.
• The training will first map the chords using the music21 framework resulting in a data structure of a list of chords.
• The data will be mapped to a String format and converted to Integer for working with Keras.
• When the training is done this will be converted back using music21 and streamed into a MIDI file.

How To Use

1. Place MIDI in folder
2. Run main.py
3. Run generate.py

Tools

Data

• The training data consisted of a large bank of chord progression MIDI files made by the company 'Unison' which is listed in the project.

Software

• Visual Studio Code

Frameworks

• Music21
• Keras
• Flask

iOS App

• iOS mobile application frontend: https://github.com/ThaDuyx/Scaile

Supporting Files

• MIDI files (.mid)