Unsync audio when transcoding to HLS multi-bitrate

[LifelessMuffin]

Hi everyone,
I got a small noticeable audio delay when transcoding to MBR HLS a high-bitrate HLS input using the %ffmpeg encoder. I'm wondering if maybe there is a way to use -vsync -1 or any other similar option/workaround.

This is my script:

input = input.http(self_sync=true,max_buffer=30.0,"http://192.168.12.25:8060/hls/2/2.m3u8")

high = %ffmpeg(format="mpegts",fflags="-autobsf+igndts+genpts",avoid_negative_ts="make_zero",%video(codec="libx264",pixel_format="yuv420p",width=1280, height=720,b="2000k",r=25,g=100,preset="medium",profile="baseline","x264-params"="scenecut=0:open_gop=0:min-keyint=100:keyint=100"), %audio(codec="aac",channels=2,ar=48000,b="192k"))

mid = %ffmpeg(format="mpegts",fflags="-autobsf+igndts+genpts",avoid_negative_ts="make_zero",%video(codec="libx264",pixel_format="yuv420p",width=640, height=360,b="1000k",r=25,g=100,preset="medium",profile="baseline","x264-params"="scenecut=0:open_gop=0:min-keyint=100:keyint=100"), %audio(codec="aac",channels=2,ar=48000,b="128k"))

low = %ffmpeg(format="mpegts",fflags="-autobsf+igndts+genpts",avoid_negative_ts="make_zero",%video(codec="libx264",pixel_format="yuv420p",width=640, height=360,b="500k",r=25,g=100,preset="medium",profile="baseline","x264-params"="scenecut=0:open_gop=0:min-keyint=100:keyint=100"),%audio(codec="aac",channels=2,ar=48000,b="96k"))


streams = [("high",high),("mid",mid),("low",low)]

def segment_name(~position,~extname,stream_name) =
  timestamp = int_of_float(time())
  duration = 2
  "#{stream_name}_#{duration}_#{timestamp}_#{position}.#{extname}"
end

output.file.hls(
                playlist="index.m3u8",
                fallible=true,
                segments=4,
                segment_duration=4.,
                temp_dir="/home/test/hls/tmp",
                segments_overhead=8,
                segment_name=segment_name,
                "/home/test/hls",
                streams,
                input
               )

Thank you in advance

EDIT:
I'm also not sure if perhaps it is caused by this. If perhaps there is a way to drop the video.text (it is an output from a DVB-S2 tv box, so I assume is teletext)

2024/04/03 11:57:40 [/home/test/hls/index_m3u8:3] Content type is {audio=pcm(stereo),video=canvas}.
2024/04/03 11:57:40 [video.converter:3] Using preferred video converter: ffmpeg.
2024/04/03 11:57:40 [audio.converter:3] Using samplerate converter: libsamplerate.
2024/04/03 11:57:40 [video.text:3] Using sdl implementation
2024/04/03 11:57:40 [clock.main:3] Streaming loop starts in auto-sync mode
2024/04/03 11:57:40 [clock.main:3] Delegating synchronization to CPU clock
[NULL @ 0x72884e0d7500] non-existing PPS 0 referenced
[h264 @ 0x72884e0da700] non-existing PPS 0 referenced
[h264 @ 0x72884e0da700] decode_slice_header error
[h264 @ 0x72884e0da700] no frame!
[NULL @ 0x72884e0d7500] non-existing PPS 0 referenced
[h264 @ 0x72884e0da700] non-existing PPS 0 referenced
[h264 @ 0x72884e0da700] decode_slice_header error
[h264 @ 0x72884e0da700] no frame!
....
[h264 @ 0x72884e0da700] mmco: unref short failure
[h264 @ 0x72884e0da700] mmco: unref short failure
[h264 @ 0x72884e0da700] number of reference frames (0+5) exceeds max (4; probably corrupt input), discarding one
[h264 @ 0x72884e0da700] mmco: unref short failure
[h264 @ 0x72884e0da700] mmco: unref short failure
[h264 @ 0x72884e0da700] number of reference frames (0+5) exceeds max (4; probably corrupt input), discarding one
[h264 @ 0x72884e0da700] Increasing reorder buffer to 2
[hls @ 0x72884e022000] Could not find codec parameters for stream 4 (Unknown: none ([5][0][0][0] / 0x0005)): unknown codec
Consider increasing the value for the 'analyzeduration' (0) and 'probesize' (5000000) options
2024/04/03 11:57:42 [decoder.ffmpeg:3] Requested content-type for "http://192.168.12.25:8060/hls/2/2.m3u8": {audio=pcm(stereo),video=canvas}
2024/04/03 11:57:42 [decoder.ffmpeg:3] FFmpeg recognizes "http://192.168.12.25:8060/hls/2/2.m3u8" as audio_2: {codec: mp2, 48000Hz, 2 channel(s)}, audio: {codec: ac3, 48000Hz, 6 channel(s)}, video: {codec: h264, 1920x1080, yuv420p}

[toots]

Hi!

With a simple script like that we should be pretty close to the native ffmpeg capabilities.

Do you also observe a delay when using the FFmpeg CLI?

[LifelessMuffin]

Thank you for your fast reply!
Yes it looks 100% related to FFmpeg, I get the same output and audio delay when running the CLI command version

ffmpeg -fflags -autobsf+igndts+genpts -i http://192.168.12.25:8060/hls/2/2.m3u8 -deinterlace -c:v libx264 -r 25 -g 100 -x264-params scenecut=0:open_gop=0:min-keyint=100:keyint=100 -c:a aac -f mpegts - | ffplay -

[toots]

Ok. Then you should be able to use any ffmpeg option. I'd say, you should first see about fixing it with ffmpeg CLI options then, if you can't use them in liquidsoap we can see about adding/fixing support for them.

[LifelessMuffin]

I sort of found it out (hopefully haha), now I'm a bit stuck in understanding how to integrate inside this Liquidsoap script the FFmpeg bitstream filter -bsf:v setts=pts=DTS+30000 (only to the video track)

I see that for example this exists
input = ffmpeg.filter.bitstream.aac_adtstoasc(input)
but how to pass the settings setts=pts=DTS+30000 ?

ffmpeg -i http://192.168.12.25:8060/hls/2/2.m3u8 -deinterlace -bsf:v setts=pts=DTS+30000 -c:v libx264 -r 25 -g 100 -x264-params no-scenecut=1:open_gop=0:min-keyint=100:keyint=100 -c:a aac -f mpegts - | ffplay -