Commercial skip worked for an episode of “Two and a Half Men” but when it skipped the commercial it also skipped some of the episode as well???
Commercial Skip (CS) is notorious for this and other problems as well. Your problem is not unique.
I’ve seen where CS starts too early, starts too late, ends too late or misses a commercial entirely. CS is preformed by an algorithm and the algorithm is not perfect.
Many programs don’t even qualify for CS: Programs & Networks Ineligible/Filtered for Automatic Commercial Skip.
Being optimistic (or naive)… My hope is that since CS is a paid subscription service, is that Tablo is constantly working to improve their CS algorithm.
As a person that has attempted to write commercial detection, you need to realize that both sides “evolve”, the commercial skipper and the broadcaster. It’s in the best interest of the broadcaster to defeat any commercial skipping mechanism you might attempt.
This could be a clue. It’s almost like saying that automobile autopilots, which use ML algorithms, will never work.
Manual recordings are not eligible as they lack the guide data details required for the feature to work.
The commercial breaks for these older programs are wildly different than those on modern, prime-time programming with either very short commercials or incredibly long ‘infomercial style’ ads which are sometimes inserted in random locations in the episode. These odd ad breaks are not compatible with the current digital signal processing (DSP) algorithms and machine learning algorithms that run on the Automatic Commercial Skip processing servers.
As a 68 year old x-IT professional I can happily say that I have never thought about the algorithms used to detect commercials in a video stream but… Maybe we should consider calling it attempted CS?
Look you guys are doing a great job now that I understand that this is an emerging technology but when it is marketed as an established technology then???
Good luck and keep up the good work!
IMHO, this will always be probabilistic at best.
For those into “AI”, I can think of a million ways to make your autonomous car into a killing machine.
Your car starts as a 2,000 pound killing machine. It’s only the human brain acting as the highest form of neural network that keeps it from killing huge numbers of people.
ML is just a set of algorithms that needs to be continually taught. Unfortunately DSP algorithms are not sexy compared to other areas of ML. So they get much less development.
Now if tablo could only use the NSA DSP algorithms.
Again, it’s not that easy. Like “evil me” making sure that Telsas all crash and burn (easy), broadcasters are actively trying to prevent commercial detection.