Parallel Double Audio Fingerprinting

This paper proposes an audio fingerprinting method called parallel double fingerprinting designed to improve the integrity of detected segments for short-duration audio indexing while ensuring robustness. The method generates two sets of fingerprints that are parallel in time by segmentation and Walsh transformation, which are binary coded. The two sets of fingerprints are matched separately when indexing, and the union of matching results is used as final result. We test our method on a data set of 100 hours TV programs recording and compare it with landmark-based fingerprinting system. The experimental results show that the overall performance of parallel double fingerprinting is improved by an average of about 3% compared to the landmark-based fingerprinting. In particular, our method has significant advantages in the integrity of matching segments. We also compared matching results of double fingerprinting with single fingerprinting, and the robustness of the double fingerprinting was significantly better.