[mythtv-users] Best advert detection settings
brendan at dbinformatics.com.au
Tue Nov 23 06:32:15 UTC 2010
On 23/11/10 15:28, Sean Grieve wrote:
> I'd like to know too :) .. Also in Australia, and currently use "All
> Methods" as it seems to work the best but only comparatively.. it's
> been pretty bad as of late.
> I know the premise is a difficult one to implement but I still think
> commercial detection based on audio volume would probably work quite
> well as the ads always seem louder. If I remember correctly I read a
> posting about this a while back saying that ads tend to have a smaller
> variation in dynamics as opposed to actually being louder.
I think that's correct, the actual volume increase isn't substantial,
they increase certain frequency ranges and compress the crap out of it
I've often thought it would be good if the automated advert detection
could be accompanied and supported by a community driven manual
fine-tuning system. To elaborate;
1. TV show x gets recorded on the same channel at the same time by 100
2. A small number of people watch the show immediately, as they watch it
they manually correct the advert detection. Rather than forcing people
to use the current edit facilitly which could get pretty old fast, it
would be better to have new key shortcut that signifies advert starts /
end on the fly similar to the bookmark function.
3. Myth instantly uploads this data to a central server or PnP system.
4. Central server collates and continually refines the indexed advert
set with some clever summing, averaging etc. This could be refined over
time to prevent abuse, a user reputation system could be employed to
weight certain users higher etc.
5. Users can opt in/out permenantly or choose to be asked case by case
if they want to follow the advert detection "stream".
6. Users then simply mark the start of the show (if it failed to
auto-detect it) then the rest should be happy sailing.
I think a system along these lines can't fail, because no matter how
much stations screw with things manual marking will always prevail where
auto-detection fails to keep up. Perhaps a really clever programmer
could even work out a way where the auto-detection system could be
trained to automatically improve or channel customise its detections
following and observing manual marking system.
That's my dream, am I mad?
More information about the mythtv-users