[mythtv] New Method for Commercial Detection.

Robert Dunn anakinberrybread at paradise.net.nz
Tue May 2 09:22:05 UTC 2006

Lucas Meijer wrote:
> Hey Robert,
>> MythTV seems like the perfect platform for me to build upon, both saving 
>> me time, and giving something back to the open source community. Whilst 
>> I am not a student, I noticed you are looking for people to develop the 
>> commercial detection aspect of MythTV for the Google Summer of Code.
>> Is this something you guys would be interested in having me work on? 
>> Please get back to me with any thoughts.
> I'm wondering what you have touched upon that makes you think keeping an 
> up to date list of commercial
> keys is not going to be a big problem though.
Bah need pictures. But based on the commercials here in NZ, Here is an 
executive summary.  Assuming in the crude ASCII pictures below:
- : Program/Undetected Commercial * : Detected Commercial

I start with a hand created database (perfect keys for the week or so 
when they were made, depending on how fast
the TV companies change their ads) what I would get with perfect keys 
would be.


After a while I would get due to some commercials not being detected


now, the undetected parts in the middle of those blocks are commercials, 
so I recreate some keys over that time space. Adding them to the 
database, removing keys which haven't been detected in a few 
weeks/months to keep the database as small as possible. Coupled with the 
possibility of on line collaboration with other users from your country, 
and other simpler techniques such as the 'volume' detection. I believe 
maintaining an update key database should be manageable.

I was planning on doing solely mono-audio fingerprinting, and as 
discussed in the PDF I linked to before, a low sample rate (5-8KHz) can 
be used, so I imagine I can get it to be relatively low processor time 
as the accuracy discussed in the article is not required here (this was 
part of my experiments on the DSP). Additional video detection would be 
later added, if the audio was not successful enough. The biggest Issue I 
see is searching a large database of keys for the detection in real time.

Hope that helped,



More information about the mythtv-dev mailing list