[Ffmpeg-devel] seek with duration for cutting: broken or changed?

[Víctor Paesa]

> Michael Niedermayer wrote:
>> Hi
>>
>> On Mon, Oct 30, 2006 at 05:37:25PM -0500, Cyrus A wrote:
>>
>>> Michael Niedermayer wrote:
>>>
>>>> Hi
>>>>
>>>> On Mon, Oct 30, 2006 at 12:55:16PM -0500, Cyrus A wrote:
>>>>
>>>>
>>>>> The following command used to work to cut a specific segment out of a
>>>>> video file on some FC3 machines I'm using with an old version of
>>>>> ffmpeg
>>>>> (still does work):
>>>>>
>>>>> ffmpeg -acodec copy -vcodec copy -i fullvideo.avi -ss 60 -t 120
>>>>> choppedvideo.avi
>>>>>
>>>>> But it doesn't work anymore with Fedora Core 5 on a 64-bit machine
>>>>> running today's SVN copy of ffmpeg. (Also doesn't work with freshrpms
>>>>> latest ffmpeg rpm, either.) I believe what it's doing is seeking, but
>>>>> the duration is completely ignored. It just returns the rest of the
>>>>> file
>>>>>
>>>> >from where the seek took it. The resulting file is also a little
>>>>
>>>>> "chirpy"  -- the audio video sync is messed up somehow.
>>>>>
>>>>>
>>>> could you find out when this problem was introduced into svn (see
>>>> binary search and svn up -r)
>>>>
>>>> [...]
>>>>
>>>>
>>>>
>>> Found it. The seek + duration option worked normally through revision
>>> 4526 and then stopped working in 4527 and every other revision
>>> thereafter. It was August 15th, 2005.
>>>
>>> The change notes read:
>>>
>>> parse pict_type for streams in avi
>>> fix mpeg4 parser so it outputs te pict_type
>>> support header only parseing without repacking
>>>
>>> That's interesting since I'm using mpeg4 for my vcodec and producing an
>>> AVI file.
>>>
>>> Man, binary searching like that is not fun. Please let me know what you
>>> think.
>>>
>>
>> i think ive just fixed it ...
>>
>> [...]
>>
> Ok, that helped. The "ffmpeg -i input.avi -acodec copy -vcodec copy -ss
> 5 -t 10 output.avi" command no longer churns as before. The cut is very
> quick and look as if it worked. But it doesn't. Here are the various
> commands I used for some tests: ("old" means 0.4.8 ffmpeg-era on 32-bit
> machines):
>
> 6849 rec: ffmpeg -i /dev/video0 -async 1 -y -vcodec mpeg4 -acodec mp3 -t
> 30 -b 1750k -ab 192 -ar 44100 -s 480x384 test.avi
> old rec: ffmpeg -r ntsc -vd /dev/video0 -ad /dev/audio -vcodec mpeg4
> -acodec mp3 -y -s 480x384 -b 1750 -t 30 test.avi
>
> 6849 cut: ffmpeg -i test.avi -acodec copy -vcodec copy -y -ss 5 -t 20
> test_cut.avi
> old cut: ffmpeg -i test.avi -acodec copy -vcodec copy -y -ss 5 -t 20
> test_cut.avi
>
> 6849 cut reenc: ffmpeg -i test.avi -acodec mp3 -vcodec mpeg4 -y -ss 5 -t
> 20 test_cut.avi
> old cut reenc: ffmpeg -i test.avi -acodec mp3 -vcodec mpeg4 -y -ss 5 -t
> 20 test_cut.avi
>
> Here is the 2x3 results table:
>                oldcopy            6849copy       6849reenc    <-- cut
> with this version
> old          OK                  aud b4 vid      no audio at all
> 6849     aud b4 vid        *aud b4 vid*      OK
> ^---- recorded with this version
>
> *= this is the one I need to work
>
> So if I record on the old version and copy-cut with the old version, it
> works. Ditto the 6849 revision when reencode cut with -acodec mp3 and
> vcodec mpeg4. Problem is, the old version won't compile on my 64 bit
> machine under FC5 (common.h: incompat element type) and the new version
> won't do a copy cut. A reencode cut doesn't help me because it takes
> forever for big files. These cuts have to be quick for web application
> use.
>
> An interesting note I found was that for those that have "audio before
> video" the audio is exactly 5 seconds before the video. So, I tried -ss
> 2 and, whammo, the audio was only 2 seconds before the video. Likewise,
> -ss 8 made the audio 8 seconds before the video. Obviously, this is the
> problem.
>
> If anyone can help me get syncd files from a "cut" process, I would
> really really appreciate it.
>

Parameters in FFmpeg are applied in the order they appear,
I believe you should set -ss 5 -t 20 before setting your input file:

ffmpeg -ss 5 -t 20 -i test.avi \
  -acodec copy -vcodec copy -y test_cut.avi

Regards,
V?ctor