CoffeMIDI

A MIDI Content Based Recomandation System

Website

http://coffemidi.com.ar/

Author:

Patricio Guinle

Motivation

To create a music search and recommendation system based on the similarity of musical characteristics that can be parameterized.

Main Notebooks

• HOST: (Notebook)
• CLIENT: (Notebook)
• DATASET GENERATOR: (Notebook)

Core Functions

• COMMONS: (common.py)

Dataset

Contains over 500 parameters from almost 90k MIDI files divided in:

• instrumentation
• drums rythm
• instruments rythm
• tonal range
• dynamics
• notes duration
• theme duration
• simult. notes
• tempo
• harmony

How to use

• Clone the repository with git clone https://github.com/PatricioGuinle/CoffeMIDI.git
• Install the requierements from requirements.txt
• Optional: Replace the example MIDI files with your own Dataset into FULL_MIDI path and run the dataset generator
• write the command line python app.py
• Open the front view in browser /Front/front.html
• You're ready to play

Web APP Usage

• Use the box to search by Artist, Music Genre or Theme name:

• Pick a Song between the results list:

• Start playing with some musical parameters to discover how recomendations begin to be adjusted by them:

In the second column of the grid result you can find the Cosine Similarity value between the picked song and the closest ones in the Database. You can Also click on the 'Search' button to change the picked theme.

• Click listen the results in order to evaluate the quality of recomendations:

You can also open a piano roll and look at the MIDI notes while playing !

Hope you enjoy it !!