[mythtv] A thought on Commercial Flagging -- Long (And rambling)

[Robert Johnston]

On 06/08/05, Jay Jarvinen <jay-lists at 3pound.com> wrote:
> On Fri, 5 Aug 2005 23:52:04 -0600
> Robert Johnston <anaerin at gmail.com> wrote:
> 
> ...
> > Anyways, How about we take the principals of Neural Networks or
> > Bayesian Filtering and apply it to Myth Commercial Flagging.
> ...
> > We can then (optionally) take these "Learned Rules" and send them
> > (anonymously) back to a central database, so every myth box
> > contributes and the whole Myth community acts as a very big (and very
> > intelligent) Distributed Neural Net.
> ...
> > be done once, by the developer of the system, and his "Base Data" can
> > be used as a starting-point for ad-detection on all other systems.
> 
> I'm following, but I'm not sure you've stated what data is used to
> create these rules. Based on your scenario where 'user hit rewind after
> auto-skip', are you suggesting time as a basis?

I'm saying that the commercial flagger takes feeds from Video an Audio
to determine commercials. It then watches the user's reactions to it's
decisions during playback to refine and determine wether these were
correct or not.

So, the commercial flagger makes it's determination and saves it.
Then, when the user plays back the file, Myth watches what the user
does. If the user asks for a commercial jump early (By pressing the
"Skip Commercials" button), Myth notes that the commercials REALLY
started at that point, and sends the relevant portion back to the
Commercial Flagger to teach it this commercial pattern. If the user
then fast-forwards from the end of the skip-point, Myth notes that it
wasn't correct and sends the fast-forwarded part to the commercial
flagger to train THAT as commercial. If the user rewinds from the
end-point that Myth has defined, Myth takes note and sends the part
the user rewound over back to the commercial flagger to teach it that
that part is NOT commercial.

> AFA brainstorming, I submit that the majority of commercials have a
> higher words-per-minute ratio than most programs (News or daytime
> talk-shows might be an exception).
> 
> Perhaps http://aubio.piem.org/ or something similar, could be used to
> map out a rough words-per-minute count.
> 
> Pitch detection is another idea, odds are that the voices/sounds in
> commercials are dissimilar to the program being recorded. Perhaps
> pitch-variation is a better term for what I'm suggesting.
> 
> Similarly, like blank frames detection, blank frames + audio-silence
> seems logical.

Also, Commercials tend to be Compressed more, making for a lower
dynamic range, and a "Louder" appearing sound (To grab your
attention).

As another point (And this is an "Enhancement"), if the flagging gets
good enough, it could be tied into Live TV, so when the flagger says
there is a commercial, Myth's Live TV "Dims" the picture and drops the
volume, until the commercial is ended, to counteract some of the
brashness and brightness of the commercials.

> Regarding a 'central database' for X, that's twice in two days now --
> see: http://www.gossamer-threads.com/lists/mythtv/users/143525

Twice, for two different reasons.

> BTW - I'm quite satisfied with the current state of commercial flagging,
> don't get me wrong. If only there were a max_commercial_length setting,
> I'd manually intervene far less often.

Commercial flagging is good, but not perfect. It still has problems
with "Law & Order" on Bravo, mainly because Bravo doesn't use a
on-screen logo.

> p.s. I'd send this to mythtv-brainstorming, but this thread is already
> here. Apologies, if you consider it noise.

There *is* a mythtv-brainstorming?
-- 
Robert "Anaerin" Johnston