Classification of music type by a multilayer neural network
Benyamin Matityaho, Miriam Furst · The Journal of the Acoustical Society of America · 1994
Western people who are exposed to different types music, such as pop or classic, can easily distinguish between them after listening to any composition for a few seconds. It is still unclear what the features are in the music that allow people this quick recognition. In the present study a model is proposed that distinguishes between two classes of music, pop and classic. The model is a decision making system that was implemented by integrating outputs of a multilayer neural network (NN). The input nodes to the NN were obtained by a preprocessing algorithm that included the following steps: (1) Dividing the musical composition to intervals of 16.8 ms; (2) applying spectral analysis on each interval; (3) combining the spectral components of T successive intervals into F divisions, which were obtained by dividing the logarithmic of the audiometric frequency range into equal F parts. As a result of the preprocessing algorithm interval of T*16.8 ms of musical signal was presented by F*T input values to the NN. Our results show that it is enough to train the NN with a group of short intervals (i.e., 0.3 s), and to represent its spectrum by about 20 values in order to obtain a 100% success in distinguishing between two types of music.