Techniques segmenting ordered information such as audio, video and text
are provided by windowing and parameterizing an ordered information stream and
storing of the parameterized and windowed information into a two-dimensional representation
such as a matrix. The similarity between the parameter vectors is determined and
an orthogonal matrix decomposition such as singular value decomposition is applied
to the similarity matrix. The singular values or eigenvalues of the resulting decomposition
indicate major components or segments of the ordered information. The boundaries
of the major components may be determined using the determined singular vectors
to provide, for example, smart cut-and-paste of ordered information in which boundaries
are automatically identified by the singular vectors; automatic categorization
and retrieval of ordered information and automatic summarization of ordered information.
