Framework for extracting multiple-resolution semantics in composite media content analysis

A non-transitory computer readable medium embodying instructions executable by a processor to perform a method for content analysis comprising:

• receiving composite media content;
• identifying at least one speech segment based from the composite media content;
• tracking a speech flow pattern with a state transition machine through a plurality of speech segments of the composite media content;
• extracting scenes from the composite media content based on the state transitions in the speech flow pattern, wherein each scene comprises a plurality of speech segments;
• and
• classifying the scenes as one of discussion content or narration content according to the speech flow pattern;
• dividing a plurality of narration-classified speech segments according to corresponding image content into a first set of homogeneous segments of the composite media content, wherein the first set of homogeneous segments each have homogeneous content;
• and
• dividing a plurality of discussion-classified speech segments according to feature space associations into a second set of homogeneous segments of the composite media content, wherein the second set of homogeneous segments each have homogeneous content.