Automatic Chord Estimation (ACE) is a central task in Music Information Retrieval. Generally, audio files are parsed into chroma-based features for further processing in order to estimate the chord being played. Much work has been done to improve the estimation algorithm by means of statistical models for chroma vector transitions, but not as much attention has been given to the loudness model during the feature extraction stage. In this paper we evaluate the effect on chord-recognition accuracy due to the use of various nonlinear transformations and loudness weightings applied to the power spectrum that is "folded" to form the chromagram in which chords are detected. Nonlinear spectral transformations included square-root magnitude, magnitude, magnitude-squared (power spectrum), and dB magnitude. Weightings included A-weighted dB and Gaussian-weighted magnitude.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=17442
(537KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
