Idea: Auto commercial removal WITHOUT file processing (e.g. Comskip)

[peter_h]

Possible? Maybe - with enough participation.

Here's the big idea:

Have HTPCs:

1) record, during live viewing, or recorded playback:

• whenever users press certain remote buttons, like MUTE/UNMUTE, FFWD, PAUSE, etc, that might indicate that a commercial break has just happened
  
• what channel they were watching
  
• the exact time they pressed the button (relative to the programme screening time)
  
• a unique identifier for the programme, such as the Country+Channel+Title+Description.
  

2) upload that info to a central internet server, where it is aggregated with all other users data, to produce a master schedule, including the most likely places commercials started and ended. Various factors could weight the results, e.g. MUTE buttons being more likely to indicate the start of a commercial; or somehow "learning" which users are more accurate than others.

3) download that data and use it instead of, or alongside, Comskip-type file-scanning software.

4) sync'd to the same clock.

The more people take part, the more accurate the data becomes.

And once the server is established, there's probably no reason why it couldn't be used for downloading/uploading schedules too.

I imagine it would be per-country based.

Unfortunately, I'm in no position to help implement any of this. But I thought I'd throw the idea out there, in case anyone wants to pick it up.

[whurlston]

the exact time they pressed the button (relative to the programme screening time)

Not only would everyone's clocks have to be synced, their pre padding would have to be as well. Pre padding can be affected by other conditions as well, like a tuner not becoming available until it's post padding period ends.

[peter_h]

I posted the same idea over at the MediaPortal forums, to see what people there had to say: http://forum.team-mediaportal.com/improv...kip-84603/

Here's some more points to add:

• Sure, the technique would really only work with a critical mass of users. But with that volume comes the ability to detect and reject erroneous data points.
  
• I see no reason why the system couldn't be adopted by all the major PVRs, e.g. MP, GBPVR, MythTV, etc.
  
• If the system was non-anonymous, then that might deter vandalism of the data (by advertisers, say).
  
• With user accounts, some users could be considered more trustworthy than others, based on history or higher trust ratings from other users. So their data carries more weight.
  
• Maybe if someone uploaded data that was way off the "norm", then they could be flagged as possible vandals, and the value of their data deprecated.
  
• What if someone could also upload a comskip.txt profile of a show. Maybe that could be incorporated into the final result. Early on, with few users, I'd imagine this would be the most reliable source of information. So at the very least, the service would be a comskip.txt sharing mechanism. So only one person has to process the file, for the whole country. (Sure, it probably wouldn't be wise to trust just one instance).
  
• Each show's "edit list" could have an "Accuracy Rating", which, amongst other things, is based on the volume of raw data processed to produce a result. A consumer of the edit list could specify a minimum acceptance level. Not all shows need have data available. Where it is, the consumer has the choice to use it - as long as it meets their minimum accuracy rating.
  
• Time sync can easily be achieved by syncing to an Internet Time Server (NTP service).
  
• Consumers could feedback to the system about the accuracy of a show's edit list - affecting it's accuracy rating for future consumers.
  

[peter_h]

The project could be mostly coded separately from the PVR core. It could supply the PVR developers with a COM object/ .NET library/dll or suchlike, with all the details about talking to the server encapsulated. I know that both MP and GBPVR use .NET.

Then it might be as simple for the PVR developers as using the following functions:

edlObject.setCountry( country )
edlObject.setHomeServerURL( "adsbegone.com")
edlObject.setTimeServerURL( "time.windows.com") ; edlObject keeps track of it's own time, so doesn't matter if system clock is wrong

edlObject.startShow( Channel,Title,Description ) ; call every time a new show starts or channel is changed

edlObject.recordButtonPress( buttonType) ; call this function every time a button is pressed!

edlObject.endShow() ; call every time a new show ends or channel is changed!

edlObject.uploadAndFlushData()

edlObject.uploadEditList( editListType, Channel, Title, Description, TimeOfSheduledBroadcast) ; for uploading a comskip.txt file


...and for downloading an edit list + accuracyRating:

editList = editListObject.getShowData( Country,Channel,Title,TimeOfSheduledBroadcast, minimumAccuracyRating)

...plus some way of specifying callbacks for getting upload/download results asynchronously.

So really, it could be made to be minimal effort to integrate into PVRs. {Of course, I'm speaking from mostly ignorance when I say something like that! }

[Reddwarf]

Due to the fact that broadcasters very rarely start precise on schedule, it would be a guesswork anyway. Ie, Nat.geo-HD channel often start from some seconds prior to scheduled time to half a minute early. The channel TVNorge is notoriously late, everything from two minutes to 10 minutes (that's why I have to have 15 min post-padding). Plus it would be a heck of a stress to sit with the remote redy to push whenever you watch tv