Starting from radio Disk Jockeys (DJs) in the early 20th century, the role of DJs has evolved, and they are now capable of headlining festivals in large stadiums. However, despite the significance of DJs in contemporary culture, the field of Music Information Retrieval (MIR) lacks an understanding of DJ techniques. The primary reasons for this include the absence of datasets and the lack of advanced analytic tools. Therefore, this thesis proposes a dataset and analytic tools, and using those, computationally analyzes DJ techniques, with a focus on mix point selection and DJ mixing.
The proposed dataset, Raveform, comprises DJ mixes, tracks played in the mixes, and structural annotations for a subset of the tracks. Using the dataset, we develop and evaluate analytic tools for 1) metrical and functional structure analysis, 2) mix-to-track alignment, 3) mix point extraction, and 4) mixing estimation. With the built analytic tools, we provide various computational analyses of DJ techniques. Finally, we propose an interactive AI DJ demo using the structure analysis model.