Key Estimation Using Circle of Fifths

Abstract

This paper presents a novel key estimation method of sound sources based on the music theory known as "circle of fifths". We firstly overview music theory and formulate the musical key analysis by vector operations. In detail, we separate music sources into small pieces and calculate FFT-based chroma vectors. They are converted to tonality vectors and COF (circle-of-fifth) vectors are calculated from the tonality vectors, which are mapped onto the circle of fifths coordinate. As a result, each music source can be represented by traces of COF vectors, which usually stay inside a single key region on the circle of fifths. Finally, HMM is applied to the traces of COF vectors in order to detect keys and their boundaries. Experiments using music databases are also carried out.KeywordsHide Markov ModelState Transition ProbabilityMusic PiecePopular MusicMusic Information RetrievalThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.