Instrument timbre chroma contours and psycho-visual human analysis

Chroma is considered as a powerful representation that is widely used in musical information retrieval applications and assumed to convey essentially the tonal information. In this paper, a new music timbre description based only on chroma contours is investigated allowing the identification of both tonal content and particularly the timbre instrument. Instrument recognition performance obtained by an automatic system, based on popular vector features and models, is compared to psycho-visual human analysis exploiting only the chroma contours. Results are very promising and the graphical analysis demonstrates that contours are dependent of the instrument and will be useful for timbre characterization.