My experience has been CS does not work very well. It is spotty as to if it can apply CS even when 10+ Mbps of upload speed is available. Even when it does apply, it doesn’t always only skip over commercials. It will sometimes perform a CS while the program is still playing but a commercial is just about to come up. Towards the end of news program the show will frequently have a 3-5 minute segment sandwiched between two 2-3 minute ad breaks. CS frequently will skip over the entire 3-5 minute segment.
CS would have required a lot of reworking to get done right. It should have had manual training. It also is not clear if it ever attempted to leverage the close captioning track as part of commercial detection. It should have also had a training mode were customers could choose to manually designate commercials for it to learn from or a way to provide feedback to correct incorrect skips.
I personally think the development time is better spent on improving manual skipping.
I thought CS was machine learning(ML) using a DSP argorithm. As opposed to neural network ML requires that it be updated with additional datasets to address any problems.
I have had the CS for about 2 years and it is a hit and miss thing. I agree with one of the above post that I mostly have problems with the sub channels. The major networks seem to work a lot better. Personally if they get rid of the service it really want matter to me. It is not that big of a deal to fast forward through the commercials especially with the nice big thumbnails which work great and the ability to pause live tv works great. I also have a Airtv 2 for sling tv and it will not pause live tv and there are no thumbnails to see how far to fast forward. I have been using the Tablo to record all of my OTA channel programs. Airtv 2 needs to step up there game
Many/most… a lot of sub-channels have classic programming. Many of these make this list, and should be marked as such - not all apps/devices will show messages.
If this is the problem with you mostly have with sub channels… it’s by-design or known issue.
I’ve had a problem with the commercial skip being dependable, but I assumed it was because of the lousy winter weather we’ve had in the southeast that causes poor signal quality. When it works, it’s fantastic. When it doesn’t, I have to go to the manual method which is OK as long as preview is available. I’ve wondered about the server side resources needed for processing millions of recordings and wondered if processing could not be shifted to the client side. My Tablo is idle 3/4 of the day, so why couldn’t the processing be performed on my Tablo instead of on a Tablo server somewhere? Granted, processing would be slower, but who cares?