[mythtv-users] Daddy, what's a commercial

Brian Wood beww at beww.org
Fri Jan 12 19:52:12 UTC 2007


On Jan 12, 2007, at 12:37 PM, Daniel Walton wrote:

>
> On Fri, 12 Jan 2007, John Drescher wrote:
>
>>> I was initially using Blank Frame Detection.  It was very fast,  
>>> but I
>>> found the accuracy unacceptably bad.  Part of this is because I  
>>> do watch
>>> a lot of shows with fade-to-black transitions.  ("Intervention"  
>>> on A&E
>>> is a good example of a show where Blank Frame Detection fails  
>>> miserably.)
>>>
>> I have used this method for over a year now and I find that it is  
>> good
>> on some shows and bad on others. One example is a soap opera that I
>> record 5 days a week for my girlfriend requires manual editing for  
>> the
>> commercials more than 75% of the time and I also have to be  
>> careful as
>> it has been known to delete several minutes of the show. However when
>> I record a show on animal planet there usually is very little or no
>> editing necessary.
>>
>>> I switched to using "All Methods" and I've been much happier.  It's
>>> slower, though.  On my VIA M10000 system it runs barely faster than
>>> realtime, around 36 fps.  I still rarely have more than a few  
>>> shows in
>>> the job queue, though, so for my purposes it's fast enough.  (The
>>> exception is if someone accidentally leaves the frontend on the  
>>> "Watch
>>> Recordings" screen.  That little preview window consumes a  
>>> surprising
>>> amount of CPU, leaving little for background jobs.)
>>
>> I'll check into that. Thanks for the info.
>
> If you are running svn you may also want to give the Experimental  
> detection
> method a try.  It currently uses logo detection and blank frame  
> detection but
> there are additional methods in the works.


The present system is certainly good enough for my purposes. I don't  
think it will ever be 100% and we are probably reaching the point of  
"diminishing returns" in that 100% more effort will result in at best  
a percent or two of improvement in performance.

I do wonder if the audio should be taken into account as well as the  
video. If there were some way to measure the "density" of the audio  
it would almost certainly be found to be higher on commercials.

The is called "psycho-acoustic perceived loudness" by Orban and  
others, and is why a commercial can sound louder than the program  
even when your VU meter says it is the same level.

Using obnoxious behavior by broadcasters and advertisers against them  
is very satisfying. The annoying "logos" have turned out to be a  
great way to detect commercial blocks. Sort of the "Judo" approach,  
using your opponent's strength against him :-)


More information about the mythtv-users mailing list