In multimedia stream, repeated sequences, e.g., commercials, jingles, usually imply potentially significant information. Therefore, mining repeated sequence is an important approach to analyzing multimedia content. This paper reports on a robust unsupervised technique of discovering repeated sequence in audio stream. Different from former research, our approach transforms the repeated sequence detection task into a Hidden Markov Model (HMM) decoding problem in a similarity trellis. To resist the false and missing matches in real application, we present a soft definition of repeated sequence, termed as maximal loosely repeated sequence (MLRS), as the objective for detection, and use a Viterbi-like algorithm to mine all the MLRSs in the stream. In addition, we propose a novel metric to evaluate the repeated sequence detection algorithm. Experiments both on simulated data and real broadcast data demonstrate the effectiveness of our method.
Bibliographic reference. Chen, Jiansong / Zhu, Lei / Feng, Bailan / Ding, Peng / Xu, Bo (2011): "A robust approach to mining repeated sequence in audio stream", In INTERSPEECH-2011, 2277-2280.