Automatic Harmony Analysis of Jazz Audio Recordings

TitleAutomatic Harmony Analysis of Jazz Audio Recordings
Publication TypeMaster Thesis
Year of Publication2018
AuthorsEremenko, V.
AbstractThis thesis aims to develop a style specific approach to Automatic Chord Estimation and computer-aided harmony analysis for jazz audio recordings. We build a dataset of time-aligned jazz harmony transcriptions, develop an evaluation metrics which accounts for jazz specificity in assessing the quality of automatic chord transcription. Then we evaluate some existing state of the art algorithms and develop our own using beat detection, chroma features extraction, and probabilistic model as building blocks. Also, we suggest a novel way of visualizing chroma distribution based on Compositional Data Analysis techniques. The visualization allows exploring the specificity of chords rendering in general and in different jazz sub-genres. The presented work makes a step toward expanding current Music Information Retrieval (MIR) approaches for Audio Chord Estimation task, which are currently biased towards rock and pop music.
KeywordsAutomatic Chord Estimation, Chroma, Compositional Data Analysis, datasets, harmony, Jazz
Final publicationhttps://doi.org/10.5281/zenodo.1467949
intranet