To create a classifier that will automatically identify music suitable for scratch tracks in film scenes.
Scratch tracks are used in film production as a temporary draft of a film’s soundtrack. This project examines a way in which film music can be categorized based on mood to aid scratch track creation in filmmaking. Five classifiers are compared to achieve the best accuracy in predicting the correct mood classification in film music. The most accurate prediction for mood classification was presented by the support vector machine algorithm and the multilayer perceptron model in this paper. The accuracy of predicting each class differed with each classifier implemented in this project. The ‘Tense’ mood class in this paper was proven, on average, more identifiable than the other mood classes; ‘Happy’, ‘Sad’ and ‘Curious’. The paper shows how the use of unsupervised learning techniques and neural networks can improve accuracy for mood classification in film music. The results present the success of machine learning techniques for classification tasks.
This project could mainly benefit filmmakers in creating automatic scratch tracks for films, it could help film makers and sound designers find appropriate music for scenes in their films. It could work in place of the scratch track so the composers have an idea of the mood that is being conveyed and the composition development can begin earlier in the filmmaking process. It could benefit machine learning development in music classification for professional and non-professional uses. Music streaming companies such as Spotify, Youtube can benefit from new proposed techniques in classifying music on mood and emotion. This project could be developed further do help identify successful composition techniques and help benefit composers and other music creation tasks.