Carleton University - School of Computer Science Honours Project
Winter 2022
Machine Learning Assisted Dynamic Musical Synthesis
Aaron Hertner
SCS Honours Project Image
ABSTRACT
Several applications exist that allow users to dynamically create their own music using only software. These, however, normally require some knowledge on music theory, or limit the user in terms of what they are able to create. M-LADMS is a project that utilizes advanced machine learning models to create accurate musical-genre predictions given a set of five input values: ‘valence,’ ‘speechiness,’ ‘acousticness,’ ‘instrumentalness,’ and ‘liveness’. Using GUI sliders to set these inputs, users can fine tune the genre they wish to replicate. M-LADMS then uses this genre specific information to assemble songs consisting of three tracks: drums, rhythm, and melody. There are a total of seven genres that the clustering model can accurately predict, within each genre there are three drum lines, three rhythm tracks, and three melodies. This, in total, means that M-LADMS boasts a library of 189 songs that are available to its users.