Forums

Compressor-limiter detection

Started by Rob Vermeulen October 7, 2004
Folks,

I want to be able to detect commercial breaks :)
So it seems that (in the Netherlands at least) on TV, the sound of
commercials is broadcasted through some massive compressor-limiter while
"regular" broadcasts such as movies and newsbreaks have "normal" sound.
I think someone has already come up with this so I'd like to know how to be
able to detect this.

I've been thinking/hypothesizing about the best way to handle this. I tried
to measure the difference in energy levels (on average) of the sound. This
did not have the preferred result :)
Maybe there are some options in the frequency domain?

I know there were VCR's in the old days that detected the blanks (black
screens) at the beginnings and ends of commercials, but that mechanism
doesn't work anymore... they don't put in blanks anymore!
With digital broadcasting one could detect a bitrate change (commercials are
usually broadcasted in higher bitrates). Although this would be difficult
with VBR broadcasts.

So, DSP gurus... I've had some good results with you in the past. Can you
help me again this time? :-)

Cheers,

Rob Vermeulen
rvermeulen  a t  arbor-audio d o t  com


Are you limited to looking at the audio portion of the broadcast, or can you
look at the video as well?  FYI, I have a VCR that performs this feature, and it
still works fine (probably 90-95% accuracy) with today's US broadcasts.  It uses
an "off-line" process, that is, after it finishes recording, it goes back and
scans the recorded segment for commercials and marks this information on tape.
Then on playback, it skips those sections.  It is a nice because it is a
"non-destructive" system--even if it messes up, you don't lose any program
material.

I'm guessing that for a robust system, you will probably need to employ multiple
methods--looking at the audio, the video, doing some timing assuming that
commercials generally come in 30-second intervals, etc..  I don't think any one
method will be fool-proof over the wide range of TV content.

"Rob Vermeulen" <rpvermeulen@tiscali-antispam-.nl> wrote in message
news:4165862c$0$44101$5fc3050@dreader2.news.tiscali.nl...
> Folks, > > I want to be able to detect commercial breaks :) > So it seems that (in the Netherlands at least) on TV, the sound of > commercials is broadcasted through some massive compressor-limiter while > "regular" broadcasts such as movies and newsbreaks have "normal" sound. > I think someone has already come up with this so I'd like to know how to be > able to detect this. > > I've been thinking/hypothesizing about the best way to handle this. I tried > to measure the difference in energy levels (on average) of the sound. This > did not have the preferred result :) > Maybe there are some options in the frequency domain? > > I know there were VCR's in the old days that detected the blanks (black > screens) at the beginnings and ends of commercials, but that mechanism > doesn't work anymore... they don't put in blanks anymore! > With digital broadcasting one could detect a bitrate change (commercials are > usually broadcasted in higher bitrates). Although this would be difficult > with VBR broadcasts. > > So, DSP gurus... I've had some good results with you in the past. Can you > help me again this time? :-) > > Cheers, > > Rob Vermeulen > rvermeulen a t arbor-audio d o t com > >
Jon,

> It is a nice because it is a > "non-destructive" system--even if it messes up, you don't lose any program > material.
I was already thinking of doing it that way. Record it all and do a check afterwards, or at the same time, depending on the realtime aspect of the detection. Then I'd save the metadata alongside the AV content and let the player decide what to play back and what not.
> Are you limited to looking at the audio portion of the broadcast, or can
you
> look at the video as well?
Sure I can look at the video as well, I just did not have any thoughts of what kind of detection I could possibly do on it. Do you know what your VCR does? Come to think of it... all stations place their logos on the topleft or topright of the screen, except on commercial breaks. Maybe I could do some detection on overlayed images. hmmm.... Enough thoughts though. but help's appreciated! Cheers, Rob "Jon Harris" <goldentully@hotmail.com> wrote in message news:2slfc7F1mo7ogU1@uni-berlin.de...
> Are you limited to looking at the audio portion of the broadcast, or can
you
> look at the video as well? FYI, I have a VCR that performs this feature,
and it
> still works fine (probably 90-95% accuracy) with today's US broadcasts.
It uses
> an "off-line" process, that is, after it finishes recording, it goes back
and
> scans the recorded segment for commercials and marks this information on
tape.
> Then on playback, it skips those sections. It is a nice because it is a > "non-destructive" system--even if it messes up, you don't lose any program > material. > > I'm guessing that for a robust system, you will probably need to employ
multiple
> methods--looking at the audio, the video, doing some timing assuming that > commercials generally come in 30-second intervals, etc.. I don't think
any one
> method will be fool-proof over the wide range of TV content. > > "Rob Vermeulen" <rpvermeulen@tiscali-antispam-.nl> wrote in message > news:4165862c$0$44101$5fc3050@dreader2.news.tiscali.nl... > > Folks, > > > > I want to be able to detect commercial breaks :) > > So it seems that (in the Netherlands at least) on TV, the sound of > > commercials is broadcasted through some massive compressor-limiter while > > "regular" broadcasts such as movies and newsbreaks have "normal" sound. > > I think someone has already come up with this so I'd like to know how to
be
> > able to detect this. > > > > I've been thinking/hypothesizing about the best way to handle this. I
tried
> > to measure the difference in energy levels (on average) of the sound.
This
> > did not have the preferred result :) > > Maybe there are some options in the frequency domain? > > > > I know there were VCR's in the old days that detected the blanks (black > > screens) at the beginnings and ends of commercials, but that mechanism > > doesn't work anymore... they don't put in blanks anymore! > > With digital broadcasting one could detect a bitrate change (commercials
are
> > usually broadcasted in higher bitrates). Although this would be
difficult
> > with VBR broadcasts. > > > > So, DSP gurus... I've had some good results with you in the past. Can
you
> > help me again this time? :-) > > > > Cheers, > > > > Rob Vermeulen > > rvermeulen a t arbor-audio d o t com > > > > > >
"Rob Vermeulen" <rpvermeulen@tiscali-antispam-.nl> wrote in message
news:41658d90$0$44064$5fc3050@dreader2.news.tiscali.nl...
> Jon, > > > It is a nice because it is a > > "non-destructive" system--even if it messes up, you don't lose any program > > material. > > Sure I can look at the video as well, I just did not have any thoughts of > what kind of detection I could possibly do on it. > Do you know what your VCR does?
I don't know for sure, but I'm guessing it is based on detecting video and/or audio blanks.
> Come to think of it... all stations place their logos on the topleft or > topright of the screen, except on commercial breaks. Maybe I could do some > detection on overlayed images. hmmm....
That might be a good option for modern broadcasts where this is common.
> Enough thoughts though. but help's appreciated! > > Cheers, > > Rob
Rob Vermeulen wrote:

   ...

> Come to think of it... all stations place their logos on the topleft or > topright of the screen, except on commercial breaks. Maybe I could do some > detection on overlayed images. hmmm....
... Often the bottom right, too. Jerry -- When a discovery is new, people say, "It isn't true." When it becomes demonstrably true, they say, "It isn't useful." Later, when its utility is evident, they say, "So what? It's old." a paraphrase of William James &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
> Often the bottom right, too.
Ok, I'll check all corners :-)
"Rob Vermeulen" schrieb im Newsbeitrag 
> > Often the bottom right, too. > > Ok, I'll check all corners :-) >
And from time to time they change from one corner to another. Regards Martin