FINDING PERCEPTUALLY SIMILAR MUSIC WITH PRECOMPUTATION

摘要

A system is disclosed which allows people who enjoy listening to musical recordings to find new musical recordings that are similar target musical recording. The system uses a model of alikeness (105) which is based upon empirical observations of human perception of likeness between pairs of songs. The humans judge a representative sample of music (101) to build the model (105). Then other musical recordings (109) are applied to this system which extracts parameters (104) from the content of each recording and applies them to the model (105) to compute the differences between the parameters of the new recording and parameters developed for the model to place them in a data base of parameter differences (107). By allowing a user to select one of the recordings from the data base for which parameters have already been extracted, the system can efficiently work with a large number of recordings. A further efficiency improvement is achieved by precomputing differences between the target recording and other typical recordings within the data base. A large data base is efficiently managed by partitioning the data base into clusters.