In this paper we present several novel techniques for incorporating fault tolerance in content-based audio search. Our algorithms extend a recently proposed framework for fast index-based search in score-like audio material. Considering queries given as a sequence of notes and the task of matching those queries to a data base of musical tunes or melodies, we investigate possible deviations such a wrong notes, missing notes, or differences in the underlying tempo curves. It turns out that our fast index-based search methods may be quite naturally adapted to tolerate the former kinds of deviations, while the case of tempo changes requires a more careful treatment. Here, we propose a new technique for incorporating a tempo tracking mechanism into our fast search algorithms. Our methods have been successfully implemented and tested within a query-by-whistling application presented at the 2001 Internationale Funkausstellung (IFA) in Berlin, Germany. We describe this application and give an overview on our extensive tests.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.