Detection of musical notes and chords using a holistic approach

There are currently two main approaches for the recognition of chords: detecting chords from pitch class profiles, which gives information about chroma, but not height, and detecting individual notes, either by peak picking from a score function or by detecting and then canceling individual notes. We propose a new method that combines both approaches: it detects chords in a holistic way, but at the same time it gives information about the chroma and height of individual notes. The approach consists in computing scores for individual notes and chords, especially those in closed form (i.e. with small intervals between notes), and then picking the candidates with maximum score, either notes or chords. This approach, inspired by the way musicians perceive chords, as a whole and not as individual notes, avoids the iterative approach of detecting and canceling notes. The method works particularly well for notes within small intervals, which tend to be hard to detect in other approaches.