Music Signal Spotting Retrieval by a Humming Query Using Start Frame Feature Dependent Continuous Dynamic Programming.

We have developed a music retrieval method that takes a humming query and finds similar audio intervals (segments) in a music audio database. This method can also address a personally recorded video database containing melodies in its audio track. Our previous retrieving method took too much time to retrieve a segment: for example, a 60-minute database required about 10-minute computation on a personal computer. In this paper, we propose a new high-speed retrieving method, called start frame feature dependent continuous Dynamic Programming, which assumes that the pitch of the interval start point is accurate. Test results show that the proposed method reduces retrieval time to about 1/40 of present methods.