[Mythtv-soc] Commercial Flagging Update July 28 2006

[Randy Robertson]

This week I've been somewhat busy helping fix last minute issues with
our first release of our software at the Earthquake center, but that
is now out of the way. Next Sunday I'll be going home to Denver for a
few weeks, and like Nikolas I'm deciding what to do about a computer.
I'll probably just end up bringing my desktop with me or connecting
with ssh and leaving it at a friends house.  I've been continuing work
on audio analysis and have found silence to be a much better
corroborative indicator than perceived loudness/compression, I'm going
to continue looking into how best to quantify the later. As far as
processing my current idea is to just average perceived intensity over
consecutive groups of frames, and give all of those frames the average
as if it were per frame, that way I can use existing per frame data
structures. Next week and beyond I will be porting old methods into
FuzzyCommDetector.  I'll be trying to maintain the ability to choose
which methods (blank/scene-change/aspect/logo-/audio/etc) to use in
the event that some methods are harmful for some people, and still get
a good result that is weighted well.  Some combinations of course, are
unlikely- you wouldn't want only audio analysis, but keeping these
segregated will make it easier to see which elements are contributing
to a better job of detection and which if any are detracting for any
given setup.