Mixing Songs with the use of Neural Networks

Abstract

The purpose of this thesis is the design and implementation of a system that automates the process of mixing music. Widely used services such as Spotify and Apply Music provide playlists grouped by genre, date of release etc. They do not provide the ability of unifying the music and the reproduction of the songs with harmony or continuity. The same applies for the commonly used media players. The system designed receives in its input a play-list and it produces in its output the proper signals the control a dj software. The method we followed uses the detection of points in time of a song using Artificial Neural Networks. The research of the thesis is based on this problem, with the best results given by a Convolutional Neural Network, in combination with the proper data representation. The predicted points given by the A.N.N. are used by a second system that automates the basic technic that deejays use to transition from the current song playing to the next. The final conclusion, regarding the detection of points in time, is the significance of the form of the input and more importantly of the output, during the training phase, with the architecture contributing to the optimization of the system.