Black Hole
May contain traces of nut
How to abort it? I set one running on HDR1, but it's not giving any debug output - only the "please wait" spinners.
Auto Detect Ad Breaks
- Thread starter njm
- Start date
- Status
Black Hole
May contain traces of nut
Now testing on a 56m HiDef BBC2 programme to see what it makes of it (and how long it takes).
Black Hole
May contain traces of nut
Nothing listed.
Black Hole
May contain traces of nut
It looks like detectads might not have installed properly on HDR1.
ChrisDaniels
Well-Known Member
I'm getting this on all my tests
25/06/2014 23:56 - err@libsndfile error:
25/06/2014 23:56 - err@File contains data in an unknown format.
Sent from my iPhone using Tapatalk
25/06/2014 23:56 - err@libsndfile error:
25/06/2014 23:56 - err@File contains data in an unknown format.
Sent from my iPhone using Tapatalk
Black Hole
May contain traces of nut
Not good, after 7 minutes it finished with "ad break found (0 - 66)", and then failed to add any bookmarks to the recording. I guess 66 frames might be too short into the recording, and there is no surprise that it had trouble with a non-commercial service, but only 7 minutes? Am I being harsh?
Black Hole
May contain traces of nut
Arrrgghhh!!
My film test, which seemed to complete OK with a batch of adverts found, did not save the bookmarks. This is reminiscent of the problems I was having with the bookmark editor not saving the bookmarks. I have looked with the hmt utility to avoid any problems with caching.
My film test, which seemed to complete OK with a batch of adverts found, did not save the bookmarks. This is reminiscent of the problems I was having with the bookmark editor not saving the bookmarks. I have looked with the hmt utility to avoid any problems with caching.
Probably : ) It's early days yet. Play with it but don't expect it to be perfect. The web interface probably has issues to be found too.
I think the detection parameters will need to be different for HD recordings. I tried it on a copy of the Hunger Games from the other night and it found lots of clusters of silence but decided they weren't concentrated enough to constitute an ad break.
What version of ffmpeg do you have installed?
Black Hole
May contain traces of nut
I ran the film test again and it worked this time (saved the bookmarks, and in the right places). I still can't get HDR1 to process anything.
prpr
Well-Known Member
First test...
It took nearly 8 minutes to analyse a 30 minute programme; the bookmarks needed to be inverted before Crop; it removed the junk at the end of the recording (there wasn't any at the start anyway); it only removed some of the ad break.
In addition, crop gave a rather rough edit - picture glitch and the sound went silent (I know this is nothing to do with detectads).
It took nearly 8 minutes to analyse a 30 minute programme; the bookmarks needed to be inverted before Crop; it removed the junk at the end of the recording (there wasn't any at the start anyway); it only removed some of the ad break.
In addition, crop gave a rather rough edit - picture glitch and the sound went silent (I know this is nothing to do with detectads).
I got the same errors. My fmpeg is 0.10
MontysEvilTwin
Well-Known Member
I have found with crop that if you delete a portion from the start of a programme it is fine, crisp and no audio loss, but if you remove a section, sometimes the picture can be briefly blocky and the sound always drops out for a few seconds after the crop point. I don't know what would be required to improve this: Drutt may be able to comment if he sees this thread. I know that the bookmark will jump to the nearest I-frame but I find that it tends to be placed a couple of seconds later than expected: presumably this is because it is only possible to bookmark while playing, not while paused.
MontysEvilTwin
Well-Known Member
You need a later version, see here.
No, 0.10 is the right version for this at the moment. Is the recording definitely decrypted?
Black Hole
May contain traces of nut
That's similar to me - about 4x real time.
It's worse when there are existing bookmarks (eg as a result of arbookmarks).
It depends what detectads is targeted for. If we just want to skip adverts, it should only place bookmarks at the end of ad breaks. That way, when watching, when an ad break starts you just press the bookmark button to skip to the start of the next segment. On the other hand, for cropping, I can see no alternative than to invert the selection before the crop - you could say the leading bookmark should be suppressed, or bookmarks that are within a smidge of the beginning and end, but there are still likely to be aberrant examples.
An idea I have had is to make crop more intelligent: what if its initial selection is whichever results in the least amount removed (with the option to invert of course)? Regardless, I still think the detectads bookmarks are only a starting point, and need reviewing and adjustment prior to committing to crop.
I was thinking that an option to only place bookmarks at the end of adverts would be useful too. It would suit the way I watch recordings - when the adverts start I decide if it's time to make a coffee or just skip through and with a bookmark at the end I can jump then fine tune if necessary before continuing to watch.
MontysEvilTwin
Well-Known Member
Apologies, I read the earlier part of the thread and took it as meaning that a later version of ffmpeg was needed. Serves me right for commenting without full knowledge of the facts
Black Hole
May contain traces of nut
Where is the bottle-neck in this process? Would it be more efficient to strip out the audio stream into a separate file first? If it is ffmpeg taking time to report on audio signal power, would it be straightforward to compile an executable which is optimised for that one task?
- Status