Shift continuous DP: A fast matching algorithm between arbitrary parts of two time-sequence data sets

Reference-interval-free continuous DP (RIFCDP) and the closely related weighted-reduction RIFCDP are techniques that have been proposed as ways to assign labels, analyze content, and search time-series data, such as audio and video. These work by detecting similar passages in two different sets of time-series data. In this paper, the author introduces a new technique, Shift CDP, that retains performance on par with RIFCDP while dramatically reducing demands for computing resources and achieving calculation times that increase in steady proportion to data volumes. This technique takes one set of time-series data as a reference pattern, and assembles standard pattern sets of fixed length, beginning from the start of the data and shifting forward in time. Continuous DP matching is attempted against each standard-pattern unit and the results are integrated. Using conversational sound data, the author compared Shift CDP against RIFCDP and weighted-reduction CDP in terms of required computing resources and performance in detecting similar passages, and demonstrated that this is an effective technique. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(10): 43–53, 2005; Published online in Wiley InterScience (). DOI 10.1002sscj.10683