For a reliable audio fingerprinting (AFP) system for multimedia service, it is essential to make fingerprints robust to the time mismatch between live audio stream and prior recordings, as well as they should be sensitive to changes in contents for accurate discrimination. This paper presents a new AFP method using line spectral frequencies (LSFs), which are a kind of parameter that capture the underlying spectral shape: the proposed AFP method includes a new systematic scheme for the robust and discriminative fingerprint generation based on the inter-frame LSF difference and an efficient matching algorithm using the frame concentration measure based on the frame continuity property. The tests on databases containing a variety of advertisements are carried out to compare the performances of Phillips Robust Hash (PRH) and the proposed AFP. The test results demonstrate that the proposed AFP can maintain its true matched rate at over 98% even when the overlap ratio is as low as 87.5%. It can be concluded that the proposed AFP algorithm is more robust to time mismatch conditions when compared to PRH method.
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.