Abstract
In general, a CBMR (Contend-based Music Retrieval) system requires heavy computation in order to retrieve an intended song specified by the acoustic input of the user. The heavy computation is primarily spent on similarity comparison, and the computation time is proportional to the number of songs in the database. The goal of this research is to speed up the similarity comparison via paralleling processing, such that the user can obtain the comparison results in a reasonable amount of time.Our parallel version of CBMR system not only has a faster response time, but also has a better recognition rate. The implementation is based on heterogeneous platforms with different operating systems and hardware configurations. With about 13000 songs in the database, a user query of"matching-anywhere" requires about 20 seconds on a single PC with Pentium 800Mhz CPU. In contrast, our system, with 15 PCs, varying from TB 1.2GHz to K6-2 300MHz, only requires less than 2 seconds to return the search result.