A quick search algorithm for acoustic signals using histogram features - Time-series active search

Proposed here is an algorithm that offers quick detection and location of a target acoustic signal, such as a specific theme song or commercial in a l ong broadcast. With previous search techniques based on shifted matching of spectra or waveforms, searching through a long signal required enormous co mputation. On the other hand, with the proposed algorithm, considerable tim e saving is obtained by using a skipping technique in the histogram-based s earch. For example, in experiments on a workstation, when spectral features were preliminarily extracted from acoustic signals, successful search for a target signal (15 s) in a 6-h signal took only about 2.3 s. In addition, the proposed algorithm proved robust to white Gaussian noise even at a SM r atio of 20 dB. (C) 2001 Scripta Technica, Electron Comm Jpn Pt 3, 84(12): 4 0-47, 2001.