Hi everyone. In addition to this (http://www.dev-archive.com/forum/showthread.php?p=1495190#post1495190) post, I post here my latest tool for audio segmentation. This tool is used for segmenting an audio stream (like the speech-music discrimination demo) BUT, the segments are classified in 8 audio classes. Furthermore, those classes are iinspired by the content usually met in movies: apart from music and speech we define gunshots, screams etc. Therefore, it is actually a tool for automatic characterization of movies, based only on the audio information. It's been created and compiled in Matlab 7 and no code is (yet) published.
If interested, visit: http://www.di.uoa.gr/~tyiannak and then go to Multi-class Audio Segmentation Demo.
I attach a screenshot of the demo. Here, a file has been loaded, the segmentation algorithm has been executed (you can see the different segments represented by different colors). The second of the figures represents the true class labels (you can use the manual annotation tool for this purpose), and the plot on the bottom represents the segments' classification probability.

Regards,
Theodore