This paper presents a new method for a quick similarity-based search throughlong unlabeled audio streams to detect and locate audio clips provided byusers. The method involves feature-dimension reduction based on a piecewiselinear representation of a sequential feature trajectory extracted from a longaudio stream. Two techniques enable us to obtain a piecewise linearrepresentation: the dynamic segmentation of feature trajectories and thesegment-based Karhunen-L\'{o}eve (KL) transform. The proposed search methodguarantees the same search results as the search method without the proposedfeature-dimension reduction method in principle. Experiment results indicatesignificant improvements in search speed. For example the proposed methodreduced the total search time to approximately 1/12 that of previous methodsand detected queries in approximately 0.3 seconds from a 200-hour audiodatabase.
展开▼
