Indexing with musical events and its application to content-based music identification

In this paper a musical event based indexing approach is proposed and its application to content-based music identification is studied. The events, which function as term words used in text retrieval or basic speech units in speech recognition, are inferred using an unsupervised learning algorithm. Its differences with the existing methods are in that the learned low-level musicology knowledge and model selection technique are exploited to extract musical events. Our experimental analyses on a task of music identification demonstrate that the proposed indexing method is efficient, compact and robust. Using a collection of 20-second query segments on the evaluation set, the equal error rate reaches 1.57%. For applications that demand fewer false alarms, we could operate the system at a reduced false acceptance rate of 0.57% while increasing the false rejection rate to 4.58%.