Automatic Music Mastering using Deep Learning

The process of mastering music has long been a tedious component of music production. Since the early ages of recorded music, it has presented a hurdle for many musicians, requiring vast technical knowledge that many musicians do not have. As technology has improved and innovations in the industry have been made, the mastering process has become simpler and has allowed many more hobbyist musicians to produce their music, without needing the help of an expert. While it has become easier, the mastering process is still a deterrent for many would-be musicians. It is both time-consuming and expensive if outsourced to a professional producer, and it also vitally impacts the quality of the song. Deep learning can excel in taking in large amounts of data and performing tasks. It is also excelling in creative applications such as art and literature, with deep learning models creating paintings and stories. The same technology has not been widely explored for music mastering. With the use of deep learning, the music mastering process can be simplified much further, even automating it completely. This research explores this idea and some of the methods that could be used for such a task. It provides evidence that this concept is viable, results of various experiments with many deep learning model parameters, as well as topics for further exploration.