Chinese, Taiwanese, and Western Hip Hop musical pieces are clustered using timbre-based Music Information Retrieval (MIR) and machine learning (ML) algorithms. Psychoacoustically motivated algorithms extracting timbre features such as spectral centroid, roughness, sharpness, sound pressure level (SPL), flux, etc. were extracted form 38 contemporary Chinese/Taiwanese and 38 Western ’classical’ (USA, Germany, France, Great Britain) Hip Hop pieces. All features were integrated over the pieces with respect to mean and standard deviation. A Kohonen self-organizing map, as integrated in the Computational Music and Sound Archive (COMSAR[6]) and apollon[1] framework was used to train different combinations of feature vectors in their mean and standard deviation integrations. No mean was able to cluster the corpora. Still SPL standard deviation perfectly separated Chinese/Taiwanese and Western pieces. Spectral flux, sharpness, and spread standard deviation created two sub-cluster within the Western corpus, where only Western pieces had strong values there. Spectral centroid std did sub-cluster the Chinese/Taiwanese Hip Hop pieces, where again only Chinese/Taiwanese pieces had strong values. These findings point to different production, composition, or mastering strategies. E.g. the clear SPL-caused clusters point to the loudness-war of contemporary mastering, using massive compression to achieve high perceived loudness.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=21066
(347KB)
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.
