[FFmpeg-user] Generating HLS chunks on demand

[Cyril Comparon]

Hi Vincent

Thank you so much for reporting the details of your journey. There seems to
be something with the TS muxer that makes it hard to split a media into
rigorously same-length segments.

I remember that in the first tests I made before answering your first email
in this thread, I tried to use -ss and -t params with both ts and mp4
output formats. The ts segments were seemingly random length whereas the
mp4 seemed strictly of the same length. Concomitantly, it seems that HLS is
now officially compatible with mp4 segments. See
https://en.wikipedia.org/wiki/HTTP_Live_Streaming#Using_fragmented_MP4 .
Could that be a way out?

Also, care to share the ffprobe command-line or other tool you're using to
extract a compact string of video frame types (e.g. "IBBBPPBBPBPB") from a
media?

Best,
Cyril

Le ven. 27 janv. 2023 à 06:51, Vincent Deconinck <vdeconinck at gmail.com> a
écrit :

> Sorry, I got my tests mixed-up in the previous email :-(
>
> The command
> ffmpeg -ss 0  -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k
> -x264opts "keyint=25:min-keyint=25:no-scenecut" -t 4 -f mpegts
> mpegts_fixed-0.ts
> creates a TS with *no missing frames*:
> mpegts_fixed-0.ts:
> IBBBPPBBPBPBPPBPBPPBPBBPP
> IPPBBBPBBPBBBPBBBPBPBBBPP
> IBBBPBBBPBBBPBBBPBBBPBBBP
> IBBPBPBBBPBBBPBBPBBPBBBPP
>
> but yet a playlist including that segment (and subsequent segments created
> the same way) is not playable.
>
> So I think the splicing is now correct, but somehow the generated TS is not
> compatible with an HLS playback.
>
> For comparison, the first segment produced by the following command:
> ffmpeg -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k -x264opts
> "keyint=25:min-keyint=25:no-scenecut" -f hls -hls_time 4 -hls_playlist_type
> vod hls_fixed.m3u8
> has the exact same gop structure:
> hls_fixed0.ts:
> IBBBPPBBPBPBPPBPBPPBPBBPP
> IPPBBBPBBPBBBPBBBPBPBBBPP
> IBBBPBBBPBBBPBBBPBBBPBBBP
> IBBPBPBBBPBBBPBBPBBPBBBPP
> But when included in a m3u8 playlist, it plays flawlessly.
>
> The files themselves are different (different size) and a simple ffprobe on
> both returns:
> Input #0, mpegts, from 'mpegts_fixed-0.ts':
>   Duration: 00:00:04.02, start: 1.458667, bitrate: 1178 kb/s
>   Program 1
>     Metadata:
>       service_name    : Service01
>       service_provider: FFmpeg
>   Stream #0:0[0x100]: Video: h264 (High) ([27][0][0][0] / 0x001B),
> yuv420p(progressive), 1920x1080 [SAR 1:1 DAR 16:9], 25 fps, 25 tbr, 90k tbn
>   Stream #0:1[0x101](fra): Audio: aac (LC) ([15][0][0][0] / 0x000F), 48000
> Hz, stereo, fltp, 130 kb/s
>
> Input #0, mpegts, from 'hls_fixed0.ts':
>   Duration: 00:00:04.02, start: 1.458667, bitrate: 1128 kb/s
>   Program 1
>     Metadata:
>       service_name    : Service01
>       service_provider: FFmpeg
>   Stream #0:0[0x100]: Video: h264 (High) ([27][0][0][0] / 0x001B),
> yuv420p(progressive), 1920x1080 [SAR 1:1 DAR 16:9], 25 fps, 25 tbr, 90k tbn
>   Stream #0:1[0x101](fra): Audio: aac (LC) ([15][0][0][0] / 0x000F), 48000
> Hz, stereo, fltp, 130 kb/s
>
> Apart from the bitrate, I see no difference.
>
> So I believe the difference lies in the hls muxer details, but I have no
> idea how to get a single slice using the hls muxer.
>
> Any idea ?
>
> Kind regards,
>
> Vincent
>
> On Fri, Jan 27, 2023 at 1:38 PM Vincent Deconinck <vdeconinck at gmail.com>
> wrote:
>
> > Hi all,
> >
> > I tried analyzing the differences between the ts files generated "all at
> > once" (using the hls muxer) and the ones generetad individually using
> -ss,
> > -t and the mpegts muxer.
> > I extracted the gop structure of the first ts segment using ffprobe then
> > reformated it in IBBP format, and here are the results
> > -f hls (works):
> > IBBBPPBBPBPBPPBPBPPBPBBPP
> > IPPBBBPBBPBBBPBBBPBPBBBPP
> > IBBBPBBBPBBBPBBBPBBBPBBBP
> > IBBPBPBBBPBBBPBBPBBPBBBPP
> > -ss -t -f mpegts (does not work):
> > IBBBPPBBPBPBPPBPBPPBPBBPP
> > IPPBBBPBBPBBBPBBBPBPBBBPP
> > IBBBPBBBPBBBPBBBPBBBPBBBP
> > IBBPBPBBBPBBBPBBPBBP
> > (and that's it !)
> >
> > In other words, the structure looks ok but 5 frames seem to be missing
> > from the second segment ! No wonder it looks choppy...
> > And indeed, I had overlooked that a simple ffprobe on the non-working
> > segment shows a duration of 00:00:03.82 while the working one has a
> > duration of  00:00:04.02 (don't know where the 0.02 comes from).
> >
> > So it seems it all boils down to frame accurate splicing indeed.
> > Any help in that regard would be greatly appreciated...
> >
> > KR,
> >
> > Vincent
> >
> >
> > On Thu, Jan 26, 2023 at 5:42 PM Vincent Deconinck <vdeconinck at gmail.com>
> > wrote:
> >
> >> Hi,
> >>
> >> OK, I made some progress, so replying to self :
> >>
> >> So although I'm forcing keyframes, the durations are actually not 4
> >>> seconds (the EXT-X-TARGETDURATION field even says 5)
> >>> Is there another way to force all segments to be exactly the same
> >>> duration ?
> >>>
> >>
> >> After a bit of reading, the -g option only forces the *maximum* interval
> >> between keyframes, but the scene detection can trigger the creation of
> >> additional keyframes.
> >> E.g. with a -g 25 in a 25fps video, if there is a scene change at frame
> >> 12, the encoder will create a keyframe at 12, and then subsequent
> keyframes
> >> could be at e.g. 37, 62, etc. , so the last gop would end after the
> >> requested segment length.
> >>
> >> The --min-keyint 25 option didn't seem to change anything, so I finally
> >> resorted to using specific x264 options with the following command :
> >> ffmpeg -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k -x264opts
> >> "keyint=25:min-keyint=25:no-scenecut" -f hls -hls_time 4
> -hls_playlist_type
> >> vod sample_fixed.m3u8
> >> and then the segments finally are exactly 4 seconds long (and the
> >> playback is OK).
> >> Good.
> >>
> >> Now I'm back to my initial problem : Dynamically generating a playable
> >> subset of segments.
> >>
> >> But cannot achieve a playable stream (tried -s mpegts, -s segment, -s
> >> hls, to no avail). For example, i ran the following commands :
> >> ffmpeg -ss 0 -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k
> >> -x264opts "keyint=25:min-keyint=25:no-scenecut" -t 4 -f segment
> >> -segment_list sample_segg.m3u8 -segment_time 4 sample0_%%d.ts
> >> ffmpeg -ss 4 -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k
> >> -x264opts "keyint=25:min-keyint=25:no-scenecut" -t 4 -f segment
> >> -segment_list sample_segg.m3u8 -segment_time 4 sample1_%%d.ts
> >> ffmpeg -ss 8 -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k
> >> -x264opts "keyint=25:min-keyint=25:no-scenecut" -t 4 -f segment
> >> -segment_list sample_segg.m3u8 -segment_time 4 sample2_%%d.ts
> >> ffmpeg -ss 12 -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k
> >> -x264opts "keyint=25:min-keyint=25:no-scenecut" -t 4 -f segment
> >> -segment_list sample_segg.m3u8 -segment_time 4 sample3_%%d.ts
> >> ffmpeg -ss 16 -i source.ts -c:v libx264 -b:v 1000k -c:a aac -b:a 128k
> >> -x264opts "keyint=25:min-keyint=25:no-scenecut" -t 4 -f segment
> >> -segment_list sample_segg.m3u8 -segment_time 4 sample4_%%d.ts
> >> then created by hand a playlist containing the generated segments, but
> >> the playback is choppy and stops after the first segment.
> >>
> >> Any idea how to create valid segments ?
> >> Is there another option than -ss and -t to splice the stream ?
> >>
> >> Kind regards,
> >>
> >> Vincent
> >>
> >>
> >>
> >
> _______________________________________________
> ffmpeg-user mailing list
> ffmpeg-user at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-user
>
> To unsubscribe, visit link above, or email
> ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".
>