Story bound segmentation task - anybody want to share ASR/CC segmentation

["Prof. Alan Smeaton" <asmeaton@computing.dcu.ie>]

Folks

This message will be of interest to those taking part in the story bound
segmentation task in TRECVID.  As you know from the guidelines, we are
all required to do certain runs, as follows:

   1.Required: Video + Audio (no ASR/CC) 
   2.Required: Video + Audio + ASR/CC 
   3.Required: ASR/CC (no Video + Audio) 
... and so on.  The rationale behind including video/audio with and
without ASR/CC is to show how video/audio information can enhance story
segmentation algorithms.

Like some of you, we at CDVP/DCU will be taking part in this task,
benchmarking our segmenter from the Fischlar system which combines audio
and video analysis only, i.e. we don't use closed captions or automatic
speech recognition.  That means we have two choices, we can

(1) build an ASR/CC segmenter and combine with our A/V analyses or
(2) use the ASR/CC segmentation provided from some other group

Option (1) is the least attractive because we would be implementing some
basic version of text-tiling or somesuch, just in order to complete the
required runs and that would not really be exploring how video/audio
analysis can enhance story segmentation.

So, my question is to those who are
(a) doing the story segmentation task and
(b) are using an ASR/CC only segmentation and 
(c) are willing to share this with us and
(d) might use our audio/video analysis in return.

Anybody interested ?

- Alan

________________________________________________________________
Professor Alan F. Smeaton          [asmeaton-at-computing.DCU.ie]
School of Computing
Dublin City University, Glasnevin, Dublin 9, IRELAND.
Tel: +353 - 1 - 7005262    Fax: +353 - 1 - 7005442
http://www.computing.dcu.ie/~asmeaton
________________________________________________________________