Unsupervised signal segmentation based on temporal spectral clustering

This paper presents an approach for applying spectral clustering to time series data. We define a novel similarity measure based on euclidean distance and temporal proximity between vectors. This metric is useful for conditioning matrices needed to perform spectral clustering, and its application leads to the detection of abrupt changes in a sequence of vectors. It defines a temporal segmentation of the signal. When the input to the algorithm is a speech signal, we further process the segments and achieve their labeling in one of three phonetic classes: silence, consonant or vowel. When the input signal is a video stream, the algorithm detects scene changes in the sequence of images. Our results are compared against classic unsupervised and supervised techniques, and evaluated with the phonetically labeled multi-language corpus OGI-MLTS and the video database of the french video indexing campaign ARGOS.