[Ffmpeg-devel-irc] ffmpeg.log.20180818

[burek]

[00:07:51 CEST] <pi--> How can I extracted the audio channel from video without re-encoding?
[00:15:34 CEST] <Cracki> "copy" codec
[00:26:17 CEST] <teratorn> when you seek an input with -ss, do you get PTSs starting at 0 in the filter graph or are the packet timestamps untouched?
[00:31:00 CEST] <lmat> I'm running ffplay on a file, and I get  https://ptpb.pw/ttxl
[00:31:25 CEST] <lmat> I'm running pulse (I'm able to use pavucontrol to change the volume of, for instance, chromium while it's playing something.)
[00:35:24 CEST] <lmat> DOH! running as the wrong user.
[00:35:28 CEST] <lmat> Thanks guys, you're the best!!
[00:36:28 CEST] <Cracki> rubberduckers are the best
[00:37:18 CEST] <lmat> Okay, new question...
[00:37:21 CEST] <lmat> "Channel layout '5.1' with 6 channels does not match specified number of channels"
[00:37:33 CEST] <lmat> I have a 8-channel audio file (.wav; it's for identifying channels)
[00:37:40 CEST] <lmat> When I play it, well, let's take a look at the output.
[00:38:30 CEST] <lmat> https://paste.ubuntu.com/p/S3crZq7WDv/
[00:38:31 CEST] <lmat> There we go.
[00:38:53 CEST] <Cracki> so sdl can't deal with 8 channels
[00:39:36 CEST] <lmat> What's SDL?
[00:39:40 CEST] <lmat> I should be using pulse?
[00:40:59 CEST] <teratorn> for playback? no
[00:41:08 CEST] <teratorn> portaudio is what you use for playback and recording
[00:41:14 CEST] <lmat> https://i.imgur.com/TxD4V2C.png
[00:41:30 CEST] <lmat> teratorn: What's portaudio? I thought I use ALSA drivers?
[00:41:56 CEST] <lmat> That is a screenshot of pavucontrol. It suggests 8 channels?
[00:42:01 CEST] <teratorn> lmat: I dunno what you are doing. it's an audio library
[00:42:18 CEST] <lmat> teratorn: I'm trying to play an 8-channel audio file on 8 different channels...
[00:42:44 CEST] <teratorn> in code? on the command line? in a browser? what?
[00:42:52 CEST] <lmat> good question! I'm using ffplay.
[00:43:04 CEST] <lmat> Earlier, I said,  I'm running ffplay on a file, and I get  https://ptpb.pw/ttxl
[00:43:14 CEST] <teratorn> ffplay is a toy for testing purposes
[00:44:00 CEST] <lmat> teratorn: okay, I'll use something else
[00:44:01 CEST] <teratorn> lmat: try vlc or mpv
[00:44:06 CEST] <lmat> (while noting that this worked a year or two ago ^_^)
[00:44:57 CEST] <debianuser> lmat: It probably still works, and those messages could be just warnings.
[00:45:19 CEST] <lmat> yeah, VLC is doing okay.
[00:45:35 CEST] <lmat> debianuser: It doesn't work because in pavucontrol, it shows only 6 channels.
[00:45:53 CEST] <lmat> Okay, I have that audio track looping, now I can get my wiring in place.
[00:45:56 CEST] <lmat> Thanks!
[00:46:13 CEST] <debianuser> lmat: But if not - it was probably pulseaudio's module injected into alsa's playback chain that is breaking SDL_OpenAudio. You can test that by uninstalling pulseaudio and check with plain alsa (you can install pulse back after the test, of course).
[00:47:01 CEST] <debianuser> lmat: On the other hand, if vlc works fine for you - just use it and be happy. :)
[00:47:42 CEST] Action: lmat is being happy.
[11:14:36 CEST] <realies> can you overlay an audio vumeter over the video stream with ffplay?
[23:13:55 CEST] <anden> so i have a video i need to transcode into two new files: one that is rotated 90 degrees and one that is not rotated. the problem is, for some reason, the quality is completed destroyed on the rotated version. i use the -vf "transpose=1" argument. any ideas?
[23:16:14 CEST] <anden> here are the full commands i ran: https://pastebin.com/raw/A7B8LBdm
[23:17:19 CEST] <TheAMM> Someone will correct me, but if you're not aiming for a specific filesize, twopass is not worth it
[23:17:30 CEST] <furq> two-pass is worth it for vpx because its ratecontrol is terrible
[23:17:37 CEST] <furq> good prediction though
[23:17:52 CEST] <anden> TheAMM: i am actually. and after doing several experiments, these were the optimal settings i came up with
[23:17:53 CEST] <TheAMM> Literally every time I offer my help someone will tell me wrong
[23:18:01 CEST] <furq> anden: you need to transpose in both passes
[23:18:05 CEST] <furq> otherwise the first pass log is useless
[23:18:20 CEST] <anden> furq: i did that too, first, but it didn't change anything for the quality
[23:18:34 CEST] <TheAMM> furq: does this apply to CRF mode as well?
[23:18:39 CEST] <anden> does it matter where on the command line the -vf argument is specified?
[23:18:42 CEST] <furq> TheAMM: in vpx, yes
[23:18:43 CEST] <furq> anden: no
[23:21:58 CEST] <anden> so anything else i can do to get the same video quality on both files?
[23:23:07 CEST] <Cracki> encode 10 seconds and check
[23:23:17 CEST] <Cracki> you had a bug, that's why.
[23:26:26 CEST] <anden> so if you guys don't have any other suggestions, i'll just export all the frames (it's not a very long video anyway) use imagemagick to batch rotate them and encode the image sequence back into the new vide..
[23:27:20 CEST] <furq> that's not really going to make any difference
[23:28:30 CEST] <anden> so you're saying it's a problem with the algorithm? that it's not able to encode the video nearly as good when the image is rotated?
[23:28:47 CEST] <furq> i'm saying you have some other problem
[23:28:54 CEST] <furq> or there's some colossal bug with vpx, which seems unlikely
[23:29:19 CEST] <anden> anyway, being out of options for the moment, i'll proceed with imagemagick and we'll see what happens
[23:29:22 CEST] <furq> but transpose literally just rotates the decoded frames, it wouldn't do anything that exporting and rotating with another tool wouldn't do
[23:29:29 CEST] <anden> yeah, that's what i was hoping
[23:29:30 CEST] <TheAMM> on ffmpeg libvpx 2pass works kind of for me
[23:29:40 CEST] <furq> anden: are you running the commands in the order in that paste
[23:29:50 CEST] <TheAMM> On mpv the logfiles are broken for some reason I haven't figured out yet
[23:29:52 CEST] <anden> yes, first pass, then second
[23:29:56 CEST] <furq> you're not running both first passes and then both second passes or anything like that
[23:30:24 CEST] <Cracki> anden, TRY vf rotate in the first pass.
[23:30:27 CEST] <Cracki> TRY it.
[23:30:36 CEST] <Cracki> and REPORT the results
[23:30:45 CEST] <furq> he said he already tried transposing in the first pass
[23:30:49 CEST] <anden> actually, the non-rotated version was done several days ago so yeah, i shouldn't be overwriting the logs
[23:30:50 CEST] <Cracki> hm
[23:30:57 CEST] <anden> i transposed in both passes, first
[23:31:04 CEST] <anden> then only in second. i never tried transposing in only the first pass
[23:31:34 CEST] <furq> anden: if you really want to pre-rotate it then just ffmpeg -i input.mkv -vf transpose=1 -c:v rawvideo -c:a copy tmp.mkv
[23:31:42 CEST] <furq> then encode that
[23:32:08 CEST] <anden> okay, so that will intermediate encode it as uncompressed video?
[23:32:11 CEST] <furq> yeah
[23:32:14 CEST] <anden> cool
[23:32:21 CEST] <furq> i doubt it'll make any difference but it's important to prevent people from using imagemagick
[23:32:27 CEST] <anden> :D
[23:32:53 CEST] <furq> use -c:v ffv1 or something if you're short on disk space
[23:33:10 CEST] <anden> the output VP9 is less than 5MB so it won't be an issue i think
[23:38:47 CEST] <poutine> I'm trying to understand how WebVTT captioning should work with HLS live. I read the HLS specification, have googled extensively, and I see synchronization to PTS is done via the X-TIMESTAMP-MAP in the webvtt segment. What I don't understand is 1) What happens if their starting segment in the live stream does not contain that PTS? 2) Can I make segment relative just by dumping the first PTS value of the segment and aligning the timeline to that? 3) If
[23:38:47 CEST] <poutine> doing SSDAI and introducing 3rd party video via HLS discontinuity, what happens if that PTS value happens in there?
[23:43:45 CEST] <anden> furq: hate to say it, but even extracting the frames from this raw video, the quality already seems bad even before encoding to VP9
[23:44:51 CEST] <anden> hold on, i might have made a mistake.
[23:46:53 CEST] <anden> nope, i still stand by my argument. maybe there's a bug with transpose?
[23:48:07 CEST] <Cracki> what's this -speed 4 stuff anyway
[23:48:25 CEST] <Cracki> could you show us screenshots of what you call "bad"?
[23:52:10 CEST] <anden> sorry guys, the raw video is fine. my bad.
[23:52:26 CEST] <anden> i guess just encoding that should be fine now.
[23:54:09 CEST] <Cracki> what was it?
[00:00:00 CEST] --- Sun Aug 19 2018