[mythtv] Re:Alternative commercial flagging method

Liam mythtv at badlysoiled.net
Sat Jan 24 14:28:45 EST 2004


I think you're right about audio, taking scenechanges and blackframes when they happen during periods of silence should really improve accuracy. I remember seeing something years ago that said that adverts use 'condensed sound' or something like that to make them seem louder than broadcast regulations allow, I don't know if this would make them more or less detectable. I haven't looked into this much, my C isn't too hot and I've only been messing with the code Chris has in there already, I suppose it would involve a lot of root mean squaring which could caine the cpu.
The code I posted failed spectaculary on a couple of shows I recorded today, it has occured to me that by looking for blackscenes that occur regulary with a constant length could work well for some channels.


> Have you considered taking audio into account as well? I have no idea how
> hard it would be to analyze audio as well as video in the current framework
> for commercial detection, but it might be useful. In the US, at least, it's
> common for commercials to be broadcast with significantly louder audio than
> regular programming. Testing for this, and maybe testing spans between black
> frames / scene changes for reasonable "commercial" durations, could possibly
make detection more accurate. 


More information about the mythtv-dev mailing list