[clug] project, video editing,editing out ads

Scott Ferguson scott.ferguson.clug at gmail.com
Tue Apr 14 05:13:44 MDT 2015


On 14/04/15 18:46, Mike Carden wrote:
> I apologise in advance for derailing the thread and probably posting
> something that I may have mentioned here years ago. Nevertheless...


Interesting and informative detour Mike - and I'm sure it leads to
Linux, somewhere. ;)

Though it's a project that's well down the list of "things to do if I
get time" (and unlikely to ever get done) I would like to investigate
both the "perceived" loudness of television advertising - and the
difference between the number of languages/subtitling for the show and
advertising as method of auto-ad marking. Processes that are likely
simpler under Linux than any other OS.

For those that are also interested - here's a couple of things from my
notes:-
Loudness levels of ads are 'supposed' to be regulated, both here and in
the US. Television stations don't care and ignore those regulations anyway.

The method used to make ads 'seem' louder is a combination of a
maximizer and a limiting filter (e.g. Maximizer and Brickwall limiter).

There are commercial products designed to defeat the increase in
'perceived' loudness. The main one is from Dolby:-
http://www.dolby.com/us/en/technologies/dolby-volume.html (see the H2
"Solving the Volume Control Problem").

There is at least one patent on using the different limiting and
compression levels to detect ads (at least that's how a cursory read
interprets it):-
http://www.google.com.au/patents/US6714594

There are a number of home-brew projects around to limit the 'loudness'
of ads. e.g.:-
http://www.instructables.com/id/TV-Volume-Loudness-Guard-using-Arduino/
and a (very) quick think makes me suspect they could be used as a basis
for an ad detector based on an algorithm that understands that ad
'breaks' don't run longer than x, or occur more frequently than y.

Looking for the nasty tricks tv stations use to make ads appear louder
would likely fail as a sole means of detecting ads as some ads are
deliberately *not* loud.

Never-the-less it'll still be interesting to manually tag ads in a
recording then examine the same point in Audacity.
=======================
Perhaps an Open Source database of annoying advertising fingerprints a
la EchoPrint/MusicBrainz??
Likewise for station logos and those annoying screen in screen
advertising scrollers??
=======================

I did get some detailed information from someone who worked in
commercial television about how the ads are inserted and the reference
tones used to set the audio levels - but I don't know how widespread
that particular practice is; whether the knowledge would be useful for
detecting ads; or really understanding what he was talking about -
though apparently the TV station does deliberately flaunt the
regulations and some advertisers insist on loud volume (one demand is to
open the ad quietly so people will forget to mute the ad break), and
other advertisers specifically request the audio levels aren't boosted.


Thanks to everyone for their input.

Kind regards




> 
> Back in the 90s I worked for a company in Sydney where we had the
> _opposite_ problem; we wanted to capture the advertising and drop the
> programme content.

I worked for the other side (Media Monitors), regrettably I never
investigated how they stripped the ads from the content which was then
reviewed and tagged for companies that collect references to particular
subjects via subscription.

> 
> The company was Tart Research and the business model was one where we
> would record TV stations all over Australia 24/7 on time-lapse VCRs. The
> recordings were passed to our cube farm where rows of data entry
> operators with headphones, TV monitors and dumb terminals would fast
> forward through TV shows to get to the adverts. Each ad would have its
> metadata manually keyed in to our database, then the operator would skip
> past the programme to find the next ad break and do it again.
> 
> The resulting data informed reports that were sold to big advertisers
> for $$$.
> 
> If ${MILLION_DOLLAR_COMPANY} had paid ${SHONKY_AD_AGENCY} a bajillion
> dollars for a campaign that promised X time slots of Y duration on Z
> broadcasters, our reports could tell them whether ${SHONKY_AD_AGENCY}
> had delivered or just taken the money and run.
> 
> In the advertising industry of the 80s and 90s, guess what the common
> outcome was?
> 
> Anyway, it was a fun business to be in and we tried all sorts of
> (analogue) ways to auto-detect adverts without a lot of success. People
> watching screens did a pretty good job.

Good point!

Maybe Fivers or one of those other online <del>slave labour</del>
lowest-price task markets could be the basis for a solution to the
"strip the ads from my recording". I'd probably pay $10 a year for the
ad markers in 6 hour old television. Given that ad placement in content
is where the main movie advertising dollar is I wonder how much money
the TV stations would consider investing in trying to defeat such a program?


Kind regards


> 
> -- 
> crash
> 



More information about the linux mailing list