Automatic-Music-Transcription-using-Deep-Learning

Sbonelo Mdluli and Moshekwa Malatji

The repo contains the implemenation of an automatic music transcription application. Automatic music transcription is the process of converting music signal into musical notes. The application transcribes 2 instruments which are:

Piano - channel 0
Drums - channel 9

Representation

The application UI is developed using PyQt5. Which can be installed using the command below.

 pip install PyQt5

Digital Signal Processing

The audio file are processed using variable Q-Transform. Digial signal processing is done using Librosa.

 pip install librosa

Machine Learning

The instrument classification model in made using Keras which can be installed using.

 pip3 install --user --upgrade tensorflow 
 pip install Keras

Data preparation

The ground truth data is generated using MIDI format. We use open source MIDI parser to get note data. The Midi score to note mapping for a piano is a one to one mapping E.g {72=G#6, 67=D#6, 64=C6}

Application Executable