Fast music information retrieval with indirect matching

This paper presents a fast content-based music information retrieval method. The high computational cost of similarity evaluation based on musical features between a pair of music clips is a crucial problem especially for searching large music database. To reduce the computational time in similarity evaluation process, the proposed method adopts an approach called indirect matching. In the approach, a small number of music clips called representative queries, which are randomly selected from a database, are used for fast computation. As an offline process, the similarities of each music clip in the database to the representative queries are recorded as a similarity table. In the online phase, the similarity between the actual query (the music clip given by a user) and each music clip in the database is quickly estimated by referring the similarity table. Experimental results have shown that the execution time of retrieval can be greatly reduced by the indirect matching without much deterioration of retrieval accuracy.