December 2019 - Ffmpeg-devel-irc

ffmpeg-devel.log.20191230
by burek 31 Dec '19

31 Dec '19

[00:00:00 CET] --- Mon Dec 30 2019

1 0

ffmpeg.log.20191230
by burek 31 Dec '19

31 Dec '19

[00:37:09 CET] <squaresmile> I'm trying to play criware's usm video. I used the VGMToolbox to extract the audio and video streams. The audio is HCA (not ADX) and plays fine so I'm inclined to believe the extraction was correct. [00:37:17 CET] <squaresmile> However, the audio seems to be h264 (not MPEG) and I can't get it to play or convert. Here's the ffprobe https://pastebin.com/raw/6Z9ydNSU and mediainfo https://pastebin.com/raw/RcHkFZdn [00:37:45 CET] <squaresmile> *the video [00:38:04 CET] <squaresmile> Any help to play the file would be appreciated [00:39:58 CET] <nicolas17> h264 is a video codec [00:40:38 CET] <nicolas17> oh I see your correction now [00:40:55 CET] <nicolas17> my eyes are tired enough for the day already x_x [01:12:39 CET] <cehoyos> Please provide a sample file [01:18:05 CET] <squaresmile> https://www.mediafire.com/file/px4u7so4wqpx6zj/ops00100_40534656.m2v/file [01:22:05 CET] <squaresmile> It should decode to something like this https://www.youtube.com/watch?v=7DvRo4o6rpw [01:25:38 CET] <cehoyos> This file is not playable, the extraction did not work correctly [01:40:55 CET] <squaresmile> hmm, I guess back to the extraction. Since the audio extracted alright, I thought it was just some messed up h264 proprietary encode [01:42:11 CET] <cehoyos> This may also be possible [03:04:26 CET] <tanlis> Hi all, I kicked off a download of a HLS stream(m3u8) using youtube-dl. It should of ended by now(football game), but the stream is still broadcasting and thus FFmpeg is still downloading data. What is the best way to force FFmpeg to stop download and begin converting/muxing the file into a viewable format? If I try Ctrl-C, it just results in a [03:04:26 CET] <tanlis> broken mp4.part file that doesn't seem to be viewable by anything. Running 4.2.1 of FFmpeg. Appreciate the help. [03:08:44 CET] <nicolas17> mp4 has an index that needs to be written at the end of the file [03:09:14 CET] <nicolas17> I thought Ctrl-C would let ffmpeg cleanly shutdown and do that finalization [03:09:41 CET] <nicolas17> but maybe not when it's running under youtube-dl? idk [03:10:02 CET] <tanlis> That's what I keep reading, but when I try it, the file just ends up as the mp4.part and I see something like: [03:10:03 CET] <tanlis> frame= 900 fps= 88 q=-1.0 Lsize= 16214kB time=00:00:30.04 bitrate=4420.5kbits/s speed=2.95x video:16162kB audio:20kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.196695%Exiting normally, received signal 2. [03:10:31 CET] <tanlis> And everything on the youtube-dl GitHub site, says to do ctrl-c [03:11:58 CET] <nicolas17> try running ffprobe on the mp4.part file [03:14:41 CET] <tanlis> Hrm, so maybe this will work. I do see this in the test file when I ran ffprobe. Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'giants-giants.mp4.part': [03:14:48 CET] <tanlis> And when I renamed the file to .mp4, that seemed to work [03:15:14 CET] <nicolas17> oh nice, so ffmpeg did finalize the file, just youtube-dl didn't rename it [03:15:37 CET] <nicolas17> does it play when renamed? [03:15:54 CET] <tanlis> The test file did. Haven't tried it on the full file. [03:16:35 CET] <tanlis> About to. Just doing a little more testing [03:22:33 CET] <tanlis> That worked. Phew [03:22:40 CET] <tanlis> Thanks nicolas17 [03:23:31 CET] <nicolas17> :) [04:39:30 CET] <Hello71> I'm inclined to guess that it has empty moov or something [04:43:45 CET] <nicolas17> Hello71: yeah, seems Ctrl-Cing made ffmpeg write the full moov at the end, but didn't make youtube-dl rename the file [04:43:50 CET] <nicolas17> so it was still a .part [04:50:42 CET] <Hello71> maybe [04:50:55 CET] <Hello71> but it seems more likely that the file is saved as streamable by default [04:51:06 CET] <Hello71> so you can do youtube-dl & mpv [04:51:49 CET] <nicolas17> how do you save an .mp4 as streamable? [04:51:57 CET] <nicolas17> fragments? [05:26:05 CET] <phk> hi guys, are scale_cuda and scale_npp filters still supports? because it not working for me while I have nvenc/cuvid encoder decoders support [09:52:57 CET] <OverCoder> hi [09:53:12 CET] <OverCoder> I'm trying to capture screen using x11grab but with Intel QuickSync [09:55:23 CET] <OverCoder> this totally works https://storage.overcoder.dev/irc/614e19c44ae0b3b4/image.png [09:55:55 CET] <OverCoder> er, I meant this https://storage.overcoder.dev/irc/aa053bccf55fcff4/image.png does work [09:56:13 CET] <OverCoder> but as soon as I switch h264 to h264_qsv, I get this: https://storage.overcoder.dev/irc/3175f1b8be0c1964/image.png [09:56:49 CET] <OverCoder> I tried going parameter by parameter migration a copy pasted ffmpeg encoding command from docs to work with x11grab - apparently I can't use QSV at all when I use x11grab? or am I missing something? [09:56:52 CET] <OverCoder> For the matter, I use vaapi [14:02:16 CET] <Etua> Hello, I'd like to write simple C program that would sort audio files using AcoustID/MusicBrainz database. AcoustID has Chromaprint - a C library which can tag already decoded audio streams, but I think that instead of externally decoding files I could wrap FFmpeg to decode files using demuxer and use Chromaprint muxer to tag them with AcoustID ID. I am not really familiar with FFmpeg architecture so I am not sure whether I get it right. [14:37:43 CET] <vlt> durandal_1707: Thank you! Using "fillborders=bottom=2" *before* "perspective=..." worked great :) [14:41:46 CET] <vlt> How can I get a clip of a specific duration from a single image? I tried `ffmpeg -f image2 -i still.png -r 25 -t 5 ...` but get an output with "Duration: 00:00:00.04" (only one frame). What do I need to add? [14:46:36 CET] <vlt> durandal_1707: (with "mode=fixed", of course, to get a black border) [14:47:17 CET] <furq> vlt: -loop 1 -i still.png [14:51:00 CET] <vlt> furq: Yes! Thans :) [17:28:18 CET] <phk> hi guys, are scale_cuda and scale_npp filters still supports [17:28:40 CET] <phk> i dont see any of those filters in -filters list [17:34:18 CET] <Etua> Hello, I'd like to write simple C program that would sort audio files using AcoustID/MusicBrainz database. AcoustID has Chromaprint - a C library which can tag already decoded audio streams, but I think that instead of externally decoding files I could wrap FFmpeg to decode files using demuxer and use Chromaprint muxer to tag them with AcoustID ID. I am not really familiar with FFmpeg architecture so I am not sure whether I get it right. [17:39:06 CET] <durandal_1707> phk: need to compile own version [17:40:24 CET] <phk> durandal_1707: according to this https://trac.ffmpeg.org/wiki/HWAccelIntro [17:40:39 CET] <phk> we just need to install nv-codec-headers [17:40:43 CET] <phk> right? [17:41:24 CET] <phk> or still need to enable anything in ffmpeg compilation time [17:44:28 CET] <phk> and if something else i have to do thn can you please tell me more? [17:45:40 CET] <pink_mist> you always need to eanble things in compulation [17:48:27 CET] <phk> pink_mist: afaik, nvenc/nvdec are autodetected at ffmpeg configuration time [17:49:29 CET] <phk> but im afraid, im wrong if scale_npp are from different package [17:50:21 CET] <phk> because i see nvdec and nvenc in encoders/decoders list [17:50:53 CET] <phk> but dont see scale_npp or cuda_scale [17:51:48 CET] <BtbN> It's called scale_cuda [17:52:08 CET] <BtbN> And you need clang or the CUDA sdk for cuda filters to be built. [18:12:39 CET] <kepstin> Etua: such a tool is certainly possible, but you can't do it in a single pass with ffmpeg. Note also that chromaprint and acoustid are not the same thing - you need to send the chromaprint to a web service, and that responds with the acoustid to tag. [18:13:57 CET] <kepstin> Etua: You can use ffmpeg to generate the chromaprint, but while you can use ffmpeg to do the second pass to add the metadata, it might make sense to use taglib or something instead, since it can (sometimes) add the metadata in-place. [18:14:48 CET] <kepstin> Etua: also, feel free to join us in #metabrainz if you want to talk about musicbrainz tagging tools :) [18:17:19 CET] <Etua> kepstin: Thank you, I'll ask the follow-up question there. [19:33:02 CET] <phk> im getting "Impossible to convert between the formats supported by the filter 'Parsed_scale_cuda_0' and the filter 'auto_scaler_0'" error in filter parsing [19:33:16 CET] <phk> in command: ffmpeg -hwaccel cuvid -c:v mpeg2_cuvid -i udp://239.255.1.10:1234 -vf scale_cuda=w=1920:h=1080 -f mpegts -y scale_cuda.ts -loglevel debug [19:33:20 CET] <phk> log: https://pastebin.com/udC87Evr [19:33:21 CET] <phk> filter_log: https://pastebin.com/t5dEEZUd [19:47:05 CET] <kepstin> phk: you're using a software encoder, so you have to use the hwdownload filter to transfer the frame into cpu memory. [19:47:18 CET] <kepstin> (or switch to using a hardware encoder, i suppose) [19:52:12 CET] <phk> kepstin: thanks it working in both ways. hwupload/hwdownload and setting hwencoder [19:54:22 CET] <phk> now problem fall in mpv :( [19:54:27 CET] <phk> not working in mpv [23:14:13 CET] <iconoclasthero> [ipod @ 0x564479a4e580] Frame rate very high for a muxer not efficiently supporting it. [23:14:13 CET] <iconoclasthero> Please consider specifying a lower framerate, a different muxer or -vsync 2 [23:14:13 CET] <iconoclasthero> [libx264 @ 0x564479a4b140] height not divisible by 2 (160x137) [23:14:13 CET] <iconoclasthero> Error initializing output stream 0:0 -- Error while opening encoder for output stream #0:0 - maybe incorrect parameters such as bit_rate, rate, width or height [23:14:31 CET] <iconoclasthero> Command line: "for i in *.flac; do ffmpeg -i "$i" -acodec libfdk_aac -vbr 4 "${i%.*}.m4a"; done" [23:14:38 CET] <iconoclasthero> how do i fix this? [23:17:56 CET] <iconoclasthero> https://pastebin.com/jPG253uz [23:22:54 CET] <furq> iconoclasthero: -vn [23:23:25 CET] <furq> or actually -c copy should work for m4a [23:23:34 CET] <furq> -c:v copy rather [23:29:23 CET] <iconoclasthero> there's no video in it. ir's a flac audio... [23:41:10 CET] <pink_mist> well why are you trying to use x264 for audio?! [23:41:40 CET] <iconoclasthero> i'm not [23:42:00 CET] <pink_mist> then the output you showed us here in the channel is probably a lie [23:42:15 CET] <pink_mist> it clearly mentions libx264 [23:43:45 CET] <pink_mist> why don't you just try furq's suggestion?! [23:48:40 CET] <iconoclasthero> i did, it worked...but i'm not 100% sure it will play on the car stereo. can i drop the 0:0 stream some how? [23:51:45 CET] <iconoclasthero> and also, why would ffmpeg lie in the output? [23:51:57 CET] <iconoclasthero> it appeared to me that the video channel might have been an image? [00:00:00 CET] --- Tue Dec 31 2019

1 0

ffmpeg-devel.log.20191229
by burek 30 Dec '19

30 Dec '19

[00:00:00 CET] --- Sun Dec 29 2019

1 0

ffmpeg.log.20191229
by burek 30 Dec '19

30 Dec '19

[02:28:56 CET] <sine0> how can i record an audio stream in ffmpeg. Im thinking some kind of curl/wget pipe stdin thingy to ffmpeg. [02:48:37 CET] <nicolas17> sine0: ffmpeg can take HTTP URLs as input, no need for curl [03:23:38 CET] <DHE> it can even take HTTPS if openssl or gnutls was available at build time [04:50:40 CET] <Hello71> muh libressl [16:40:32 CET] <M6HZ> Hello, I'm trying to test the hardware acceleration of h264 encoding with h264_vaapi, but I get this error: Impossible to convert between the formats supported by the filter 'Parsed_null_0' and the filter 'auto-inserted scaler 0' [16:42:36 CET] <M6HZ> Do you know what I should do? [16:50:00 CET] <ddubya> what is the best command line to get a 0 exit status and only probe the streams? .e.g ffmpeg -to 0.01 -i <input> -f null dummy [16:50:28 CET] <ddubya> *without using ffprobe* [16:50:48 CET] <pink_mist> I was about to suggest using ffprobe ... [16:50:57 CET] <pink_mist> why don't you want to use ffprobe?? [16:57:47 CET] <ddubya> I'm doing an ffmpeg gui, I only want to require ffmpeg.exe [17:05:42 CET] <DHE> you have to be careful that a fully named "dummy" doesn't actually exist though... or add -y to that command [17:24:41 CET] <ddubya> DHE, good point, turns out "-f null" will not touch the output file [17:48:38 CET] <voices> hey, are there any good resources about using ffmpeg for HLS? [18:06:46 CET] <DHE> depends on your goal. there are a few different ways to do HLS... realtime vs VOD, and single vs multi bitrate [18:06:53 CET] <M6HZ> I still haven't solved my problem, does anybody have any idea? [18:26:34 CET] <voices> DHE: basically i want to automate converting .mkv files to 720p, have them retain their original aspect ratio, and output .ts segments [18:28:39 CET] <voices> so i think that's video on demand [18:39:29 CET] <DHE> voices: yeah that sounds right. so most likely you want: ffmpeg -i input.mkv -c:a aac -b:a 128k -ac 2 -vf scale=-2:720 [H264 options go here] -hls_time 8 -hls_list_size 0 -f hls output.m3u8 [18:39:53 CET] <DHE> hls_time is the duration of the .ts files, and what's important is selecting your keyframe interval (the -g parameter) to be suitable for the number given [18:43:54 CET] <voices> DHE: thanks. are you sure it's 2:720 ? i thought it was 1:720. [18:44:50 CET] <DHE> voices: the absolute value of the negative number indicates the multiple required. -2 means an even number must be selected for this value (because H264 in yuv420 requires it) [18:45:32 CET] <DHE> -1 will give the most correct calculation, but -2 is required for 4:2:0 colourspace [18:46:47 CET] <voices> DHE: right, yeah that was a confusing part of it. because the segment sizes vary, like so https://i.imgur.com/VykgUOB.png [18:46:50 CET] <voices> oh ok [18:47:04 CET] Action: voices reads up on 4:2:0 colourspace [18:47:32 CET] <DHE> voices: https://en.wikipedia.org/wiki/Chroma_subsampling#Sampling_systems_and_ratios pictures help [18:50:41 CET] <voices> DHE: cool. it sounds like i don't have a choice anyway [18:51:58 CET] <DHE> basically yeah. h264 supports 4:4:4 but you can't rely on players to be able to handle it [18:55:01 CET] <voices> DHE: oh, right. i think i read about that now you mention it. how does it know what size to make each segment? i expected each segment to be the same size but they're all different. and different videos, seem to be differetn again. for instance one video was divided into 19 segments, and no segment was greater than 2.5MB, but as you can see in the image i posted, those segments are around 5.0B [18:56:03 CET] <DHE> file size will vary with bitrate on a second-by-second basis, but if the durations are variable then you need to lock down you keyframe interval. [18:56:46 CET] <DHE> with libx264 you can use -x264-params no-scenecut to force exact keyframe pacing [18:56:57 CET] <voices> haha are you just making up words now? just kidding [19:19:03 CET] <voices> DHE i think the durations are the same anyway. thanks for your help. anything else i ought to know? [19:21:53 CET] <DHE> ideally you want a constrained max bitrate so that the bitrate doesn't spike up too high. HLS is designed for streaming so you should know what the max bitrate is if it's going on bandwidth constrained environments like the internet [19:22:06 CET] <DHE> if this is some kind of home LAN thing it probably doesn't matter [20:01:33 CET] <voices> DHE: so what are thew h264 options i'll need [20:03:15 CET] <DHE> unless you have specific requirements, something like: -profile:v high -crf 25 -maxrate:v 8M # adjust numbers to your needs [20:05:19 CET] <voices> DHE: i'll give it a whirl. appreciate ya [20:30:26 CET] <analogical> how do I tell ffmpeg to use a certain bitrate when I encode aac ? [20:31:36 CET] <relaxed> -c:a aac -b:a 192k [20:33:49 CET] <analogical> relaxed, Thanks. FFmpeg is a great tool but it has the worst documentation in the history of software. [20:34:31 CET] <relaxed> https://trac.ffmpeg.org/ and https://trac.ffmpeg.org/wiki/Encode/AAC [20:37:20 CET] <relaxed> once you get the hang of using basic commands the documentation is easier to search/read [20:37:53 CET] <analogical> I've been reading this https://ffmpeg.org/documentation.html and it's a nightmare [20:43:32 CET] <durandal_1707> analogical: stop shitting about documenation [20:44:39 CET] <relaxed> Yeah, they're man pages converted to web pages. I recommend focusing on the wiki and reading actual man pages when needed, also "ffmpeg -h full | less" [20:57:07 CET] <furq> you must not have used very much software [20:58:34 CET] <furq> the only major gripe i have with the docs is generic lavc options not having defaults/ranges documented per-codec [20:59:00 CET] <furq> and maybe some formats/codecs being missing from the web docs, although they're covered fine by -h [20:59:07 CET] <furq> web docs/manpages [20:59:38 CET] <furq> i'd say that's pretty good considering how colossal ffmpeg is [21:38:56 CET] <FLXX> hi [21:41:05 CET] <FLXX> can one help me with cropping please? [21:41:39 CET] <FLXX> I have a video that is has horizontally 2 black rects besides. [21:41:50 CET] <FLXX> and I want to crop them away [21:42:38 CET] <FLXX> I have 720x1280 in original [21:42:41 CET] <FLXX> and "D:\FFMPEG\ffmpeg.exe" -i "G:\3.mp4" -filter:v "crop=480:1280:0:1280" -c:a copy "G:\vivi\Tout.mp4" [21:42:45 CET] <FLXX> as commadline [21:43:09 CET] <FLXX> but anyhow it only gets squeezed horizontally and smaller [21:56:54 CET] <M6HZ> I'm trying to test the hardware acceleration of h264 encoding with h264_vaapi, but I get this error: Impossible to convert between the formats supported by the filter 'Parsed_null_0' and the filter 'auto-inserted scaler 0' [22:14:56 CET] <rascal999> hello I have mkvs which I've tried to convert using -codec copy to mp4s for playback in browser [22:16:03 CET] <rascal999> when trying to play resulting file in browser, audio doesn't work or video doesn't, what command do i need to convert (lossless as possible) and always work in browser? [22:17:18 CET] <pink_mist> depends. if the video and/or audio codec is something your browser supports and mp4 supports it too, copy is the best option [22:17:33 CET] <void09> any way to pass custom encoder parameters that are not available by default with ffmpeg ? [22:17:40 CET] <pink_mist> if it isn't, find what your browser does support, and use the best encoding options for that codec [00:00:00 CET] --- Mon Dec 30 2019

1 0

ffmpeg-devel.log.20191228
by burek 29 Dec '19

29 Dec '19

[00:00:00 CET] --- Sat Dec 28 2019

1 0

ffmpeg.log.20191228
by burek 29 Dec '19

29 Dec '19

[07:46:52 CET] <AiNA_TE> how do i set "Average quantizer" value in ffmpeg? [07:47:59 CET] <AiNA_TE> and is it known by some other name?, i don't see it in the documentation [09:37:41 CET] <celebrate> hey everyone [09:37:55 CET] <celebrate> How can i make the scaling option respect aspect ratio of the video? [09:37:56 CET] <celebrate> .outputOptions(['-s 480:600','-c:a copy', '-c:v libx264', '-preset:v medium']) [09:38:06 CET] <celebrate> i tried 480:-1 for example but that didn't work [09:38:08 CET] <celebrate> here's full my cmd [09:38:13 CET] <celebrate> ffmpeg() .input(tempFilePath) .input(watermarkFilePath) .outputOptions(['-s 480:600','-c:a copy', '-c:v libx264', '-preset:v medium']) .complexFilter([ "[0:v][1:v]overlay=W-w-20:H-h-20" ]) .format('mp4') .output(targetTempFilePath) [09:39:16 CET] <celebrate> It gave me an error when i tried: Invalid frame size: 480:-1. [09:39:28 CET] <celebrate> The top example works but doesnt respect aspect ratio [12:23:11 CET] <sopparus> hello [12:24:31 CET] <sopparus> a few years ago I used some ffmpeg or possibly mkvtools to find mkvs that kodi couldnt jump to positions in, to fix it I just remade the mkv. Anyone knows how you can figure this out with cli? [13:55:21 CET] <LFSVeteran> ffmpeg -f alsa -acodec pcm_s32le -ac 2 -ar 192k -i hw:1,0 -strict -2 -c:a vorbis -content_type audio/ogg -f ogg icecast://user:pass@127.0.0.1:8000/stream [13:55:29 CET] <LFSVeteran> how can I add buffering to this? [13:55:42 CET] <LFSVeteran> the stream is working, but faltering [14:58:49 CET] <relaxed> LFSVeteran: you probably want '-c:a libvorbis', ffmpeg has a native encoder called 'vorbis' and it sucks. [15:15:15 CET] <scramblez> Hi All, I have an mp4 with a mono aac audio stream in it. How could I map it to individual channels at a time? e.g. how could I map it to come out only on FR, or only on LFE? [15:16:36 CET] <scramblez> while the other channels stay silent/empty? [15:17:24 CET] <scramblez> Also, convert it to ac3 at the same time? [15:19:29 CET] <scramblez> Ahh ... 'channelmap' may be what I need ... [15:22:33 CET] <relaxed> scramblez: look at the channelmap filter [15:22:39 CET] <relaxed> oh, nm :) [15:23:11 CET] <scramblez> relaxed: thanks, I'm reading up on it now. :-) [15:24:25 CET] <scramblez> Hmm ... struggling with this. [15:24:51 CET] <scramblez> I have a single mono aac audio. How do I direct it with channelmap? [15:25:05 CET] <scramblez> Say to only play out of FR? [15:25:31 CET] <scramblez> What should the channelmap input be for a mono? [15:26:11 CET] <scramblez> Maybe I need to map the audio stream 0:1? [15:26:37 CET] Action: scramblez back in a while ... [15:35:14 CET] <vlt> Hello. Is there a filter for shearing video? Im looking for a quick way to move just the lower left corner of a video up by a few pixels. Any idea? [15:38:35 CET] <Hello71> don't think so. if you're just editing one video it's probably easier to just load some graphical video editor [15:40:31 CET] <vlt> I found "perspective" which almost does what I need. [15:41:23 CET] <vlt> Any idea how to fill missing parts with black background instead of repeating the pixels of the last line? [15:51:20 CET] <LFSVeteran> relaxed: ah that explains why libvorbis isn't working...it says unknown encoder [15:51:47 CET] <LFSVeteran> checking configure how to enable the system one... [15:53:55 CET] <durandal_1707> vlt: you can add fillborders filter [16:00:45 CET] <LFSVeteran> rebuilding ffmpeg with correct vorbis support [16:24:42 CET] <vlt> durandal_1707: Trying that, thanks. [16:43:10 CET] <sopparus> anyone know how to detect missing or bad index in a mkv file using ffmpeg or mkvtools? [18:35:33 CET] <LFSVeteran> hmm libvorbis won't allow 192kHz sample rate [18:35:48 CET] <LFSVeteran> while I can do with vorbis [18:36:18 CET] <BtbN> Why would you even need that high of a samplerate? [18:36:26 CET] <BtbN> Doesn't that mess with players which most likely don't support it? [18:37:15 CET] <LFSVeteran> forgot to mention it isn't just audio what I want to stream... [18:37:59 CET] <BtbN> That makes that sample rate even more questionable [18:38:12 CET] <LFSVeteran> so for normal audio, correct, 44kHz would be sufficient [18:38:35 CET] <LFSVeteran> the max freq is about 60kHz [18:38:59 CET] <LFSVeteran> which is possible with alsa and the soundcard [18:39:28 CET] <BtbN> But not with the vast majority of speakers, headphones and ears. [18:40:18 CET] <LFSVeteran> no indeed, you'll need some filtering there....that's what happens, audio gets separated from the rest [18:41:12 CET] <LFSVeteran> maybe just plain pcm_s32le? [18:49:59 CET] <furq> applying audio compression to non-audio data is generally a bad idea [18:51:00 CET] <furq> lossy compression is much more likely to lose stuff you care about and lossless compression probably won't compress it very well [18:52:56 CET] <LFSVeteran> options for non compressed codecs/formats? [00:00:00 CET] --- Sun Dec 29 2019

1 0

ffmpeg-devel.log.20191227
by burek 28 Dec '19

28 Dec '19

[00:00:00 CET] --- Fri Dec 27 2019

1 0

ffmpeg.log.20191227
by burek 28 Dec '19

28 Dec '19

[00:19:17 CET] <BtbN> kepstin, yes, it seems like the vast majority of my issues just went away by pulling a more recent version with that patch in. [14:28:38 CET] <bencc1> kepstin: I've added "-af aresample=async=1" and now I'm not getting "Queue input is backward in time" warnings in the log [16:14:08 CET] <barhom> Is cuvid (legacy) deprecated over cuda hwaccel for decoding? Are the plans to remove cuvid in the future? [17:01:10 CET] <kepstin> barhom: my impression is that the nvdec/nvenc stuff is the current recommendation [17:01:46 CET] <barhom> kepstin: i.e. cuda? [17:02:19 CET] <barhom> been trying to google the difference between cuvid and cuda but I cant find any good information on the subject [17:02:20 CET] <kepstin> i didn't say cuda, i said nvdec [17:02:39 CET] <barhom> kepstin: The question was related to "-hwaccel cuda" [17:02:54 CET] <barhom> "nvdec" is an alias of cuda afaik [17:03:33 CET] <barhom> as such I think cuda should be the right name, there are no reference to "nvdec" in "ffmpeg -hwaccels" [17:04:02 CET] <kepstin> ... what ffmpeg version do you have? [17:04:16 CET] <barhom> git-2019-12-27 [17:04:55 CET] <kepstin> weird, i wonder if it was compiled with nvdec disabled somehow [17:05:27 CET] <barhom> I can still use "-hwaccel nvdec" but its the same as "cuda" [17:05:45 CET] <barhom> Hardware acceleration methods: cuda, cuvid are the only two options [17:06:13 CET] <kepstin> oh, huh, i didn't realize that it was named cuda [17:06:29 CET] <kepstin> it's weird because all of the actual source code files are named "nvdec" :/ [17:06:53 CET] <barhom> Anyways, the question is if cuda/nvdec (same thing) is the preferred option over cuvid. From what I understand from what I have read it is. And someone mentioned that cuvid is deprecated but I cant find any source to that. [17:07:16 CET] <barhom> If someone do have a nvidia card I would love if you can try what I have written here and see if you get the same results as me: https://trac.ffmpeg.org/ticket/8442 [17:09:17 CET] <kepstin> barhom: can you reproduce only using the hardware decoder, with a software encoder? [17:09:42 CET] <barhom> kepstin: the encoder doesnt matter you can use h264_nvenc or libx264 [17:09:58 CET] <barhom> what makes or breaks it is the decoder (cuda vs cuvid) [17:10:36 CET] <kepstin> barhom: it would be good to note that in your ticket, since it's possible it could have been an interaction between the hardware decoder and encoder [17:11:50 CET] <barhom> Yeh, I'll mention it. [17:11:58 CET] <kepstin> hmm, my only pascal card with nvdec isn't in a working computer right now :/ [17:12:13 CET] <barhom> The only difference between the two commands are the decoding (before the -i) [17:17:13 CET] <kepstin> just to note that if i do try to reproduce this, my 1030 gt only has nvdec, it does not have nvenc, so I cannot test with a full hardware pipeline. and it's possible that there's a difference. [17:46:22 CET] <bencc1> kepstin:"-af aresample=async=1" 'm not getting "Queue input is backward in time" warnings in the log [17:46:34 CET] <bencc1> and there is no noticable effect on audio [17:47:14 CET] <bencc1> I've listened to mp4 before and after and I can't distinguish. no gaps or pitch issues [17:47:16 CET] <bencc1> is it safe to use? [17:48:05 CET] <bencc1> someone also suggested using "-af asetpts=N/SR/TB" [17:54:10 CET] <kepstin> i would have expected the aresample thing to cause a pitch change, since in theory it should be stretching the audio out to match the packet timestamps [17:54:31 CET] <kepstin> the asetpts would probably cause desync, but feel free to try it [17:55:50 CET] <bencc1> kepstin: maybe the pitch difference is unnoticeable because aresample is streching at most part of a sample [17:56:11 CET] <bencc1> is there audio file that I can test with that will enhance the pitch difference? [00:00:00 CET] --- Sat Dec 28 2019

1 0

ffmpeg-devel.log.20191226
by burek 27 Dec '19

27 Dec '19

[00:00:00 CET] --- Thu Dec 26 2019

1 0

ffmpeg.log.20191226
by burek 27 Dec '19

27 Dec '19

[01:11:22 CET] <johnjay> kepstin: the guy had one other video about the 480i stuff [01:11:27 CET] <johnjay> https://www.youtube.com/watch?v=1MY-jUkogbg [01:11:48 CET] <johnjay> mostly interlacing sounds like a nightmare for capturing [01:38:10 CET] <another> very interesting. thanks johnjay [01:40:34 CET] <johnjay> np. i don't know much C so i'm always looking for good articles or tutorials on video stuff [01:40:46 CET] <johnjay> otherwise i guess you could read ffmpeg src and learn everything. XD [02:20:15 CET] <analogical> I'm noticed something really strange. All incompletely downloaded .mkv files are playable but only some incompletely download .mp4 files are playable. Home comme? [02:21:14 CET] <analogical> *come [02:36:44 CET] <DHE> mp4 files need muxing options to be stream-friendly [02:37:00 CET] <DHE> and a truncated file is, like, a partial stream sorta [03:44:36 CET] <kepstin> analogical: it's basically a design flaw in the mp4 container. some initialization data the codec needs to start decoding is stored in the same place as the index [03:45:05 CET] <kepstin> analogical: and the index is variable size and can't be written until after the video's encoded, so it's normally put at the end of the file. This means the initialization data is also at the end of the file [03:45:15 CET] <kepstin> analogical: so if you're missing the end of the file, you can't decode anything [03:46:00 CET] <kepstin> analogical: ffmpeg (and other tools) can optionally do a second pass to move this data to the start of the file instead - that's what happened in the files that are playable for you. [03:47:52 CET] <analogical> interesting! [04:06:34 CET] <kepstin> this is also why you shouldn't use mp4 files for things like live screen capture - if something crashes/breaks, the resulting partial file won't be playable [04:13:40 CET] <johnjay> kepstin: that sega genesis dithering looks gorgeous doesn't it [04:14:03 CET] <johnjay> amazing what you can do with just composite video [09:16:36 CET] <bencc1> I have a/v sync issue in adobe premiere with mp4 captured using ffmpeg [09:17:04 CET] <bencc1> in "ffprobe -i cap.mp4 -show_streams" I see that [09:18:01 CET] <bencc1> The video stream start_time=0.020996 while the audio stream start_time=0 [09:18:37 CET] <bencc1> video start_pts=1290 and audio start_pts=0 [09:18:58 CET] <bencc1> video timecode=N/A [09:19:39 CET] <bencc1> should I fix this when I capture with ffmpeg? could this cause a/v sync issue in a video editor? [12:56:07 CET] <chinknot> I'm trying to use pipes to input an embedded image from an audio file into a second ffmpeg command, why is my pipe broken? https://pastebin.com/raw/iTmffq7j - the error says: [image2 @ 0x7fffcb59bac0] Could find no file with path 'pipe:0' and index in the range 0-4pipe:0: No such file or directory [14:24:08 CET] <kepstin> chinknot: the 'image2' muxer/demuxer is designed to work only with files, since it turns each frame of video into a separate file. You can try 'image2pipe' instead. [14:26:24 CET] <chinknot> thank you. it worked this time [15:41:36 CET] <bencc1> what does time_base=1/61440 in ffprobe means? [15:41:58 CET] <bencc1> I've created a 30 CFR file so I think it should be 1/60000 [15:44:21 CET] <Hello71> not sure, but I think it doesn't matter [15:44:22 CET] <Hello71> as long as the numerator is right [15:50:08 CET] <Derpton899> Hello? [15:50:12 CET] <Derpton899> Oh finally. [15:51:49 CET] <Derpton899> Hi. I have a question about ffmpeg. Is it possible to save a video capture stream directly to disk drive without encoding it? I have a USB capture card plugged into my raspberry pi 3b+, and I want to save the video footage from the capture card into my drive without encoding it as raspberry pi does not have that much power to do so. [15:55:48 CET] <BtbN> In theory, yes. In reality, your storage will be too slow to sustain that. [15:56:32 CET] <BtbN> Specially with the USB Cap card already fighting with the USB Storage about the very limited USB2 bandwidth. [15:57:29 CET] <Derpton899> I have an external HDD connected to it using USB. I have a streaming server set up on it which I use to download files using torrents and stream from rpi. It can handle around 80 mbps read/write [15:58:32 CET] <Derpton899> I think that should be sufficient? Or is it still too slow? [16:01:02 CET] <Derpton899> https://pastebin.com/raw/N93we0xd this is what I'm working with. [16:03:02 CET] <Derpton899> Is it possible to tell me what I need to do to achieve what I plan to do? [16:03:48 CET] <Derpton899> I have tried "ffmpeg -f v4l2 -input_format mjpeg -i /dev/video0 -c:v copy test.mkv" but I'm not sure if that is the correct way of doing it. [16:08:03 CET] <Hello71> don't some versions of rpi have hardware encoding [16:08:25 CET] <Hello71> knowing them it's some proprietary blob that doesn't work in ffmpeg, but still [16:09:22 CET] <Derpton899> I'm not sure if hardware encoding is available in rpi 3's. [16:19:50 CET] <wondiws> Derpton899, I'm pretty sure it is [17:18:06 CET] <bencc1> how can I ensure constant audio pkd_duration and pkt_duration_time when capturing from pulseaudio? [17:18:39 CET] <bencc1> I see in "ffprobe -show_frames" that each audio packet has different duration [17:18:44 CET] <bencc1> this is how I capture: [17:18:48 CET] <bencc1> ffmpeg -f pulse -ac 2 -ar 48000 -i default -f x11grab -framerate 30 -i :0.0 -acodec aac -b:a 96k -vcodec libx264 -pix_fmt yuv420p -preset:v veryfast -crf 23 -vf fps=30 cap.nut [19:45:17 CET] <kepstin> bencc1: i dunno what's up, aac has a fixed frame size of 1024 samples, so all the audio packets should have the same duration (excluding preroll, of course). [19:45:38 CET] <kepstin> bencc1: pulseaudio has nothing to do with it, since the audio gets re-framed before encoding to meet the encoder requirements [19:53:01 CET] <bencc1> kepstin: part of the frames: https://pastebin.com/raw/242My9tC [19:53:42 CET] <bencc1> ffmpeg -f pulse -ac 2 -ar 48000 -i default -f x11grab -framerate 30 -i :0.0 -acodec aac -b:a 96k -ar 48000 -vcodec libx264 -pix_fmt yuv420p -preset:v veryfast -crf 23 -vf fps=30 cap.nut [19:54:01 CET] <bencc1> also tried to add "-ar 48000" to -acodec but still doesn't help [19:54:30 CET] <bencc1> do I need something like "-af "aresample=async=1" ? [19:54:43 CET] <kepstin> what's the timebase? [19:54:56 CET] <kepstin> (need -show_streams in ffprobe to get that) [19:56:28 CET] <bencc1> https://pastebin.com/raw/PD76LyCz [19:56:39 CET] <bencc1> kepstin: ^^ -show_streams [19:57:55 CET] <bencc1> I also don't understand why video stream has "time_base": "1/61440" instead of "1/60000" [19:58:32 CET] <kepstin> that's an mp4 quirk, x11grab uses microsecond timebase but it gets reduced a bit to go into mp4 [19:58:46 CET] <kepstin> so yeah, audio's using sample rate timebase, that's fine. [19:59:15 CET] <kepstin> is the ffmpeg capture command running slower than realtime? [19:59:34 CET] <bencc1> I do get warnings when it runs [19:59:44 CET] <kepstin> oh, humm, you're using -vf fps=30, i have no idea why tou'd have that weird timebase [19:59:49 CET] <kepstin> either way, mp4 muxer issue :/ [20:00:12 CET] <bencc1> "-vf fps=30" ensure I have CFR for video [20:00:17 CET] <kepstin> wait, your command and that output don't match. the output is mp4, but your command has nut [20:00:19 CET] <bencc1> I think I need something similar for audio [20:00:42 CET] <bencc1> ffmpeg -i cap.mkv -c:a copy -c:v copy -chunk_duration 1000k -movflags +faststart out.mp4 [20:00:56 CET] <kepstin> and that has an mkv [20:00:57 CET] <bencc1> I capture with .nut and at the end convert to mp4 [20:01:07 CET] <bencc1> ffmpeg -i cap.nut -c:a copy -c:v copy -chunk_duration 1000k -movflags +faststart out.mp4 [20:01:09 CET] <bencc1> fixed [20:01:18 CET] <kepstin> ignore the mp4, look at the nut file you captured :/ [20:01:19 CET] <bencc1> ffmpeg -i cap.mkv -c:a copy -c:v copy -movflags +faststart out.mp4 [20:01:25 CET] <bencc1> why? [20:01:47 CET] <kepstin> because you need to narrow down the cause of the issue. [20:01:50 CET] <bencc1> I copy the audio and video streams. do you think the muxer changes frame time? [20:01:54 CET] <bencc1> ok [20:02:11 CET] <bencc1> is there a way to force constant audio frames on the output? [20:02:24 CET] <bencc1> similar to "-vf fps=30" for video? [20:02:29 CET] <kepstin> when you did the original capture, was the "speed" output of the ffmpeg stats line 1.0 ? [20:02:40 CET] <bencc1> close to 1.0 [20:03:08 CET] <bencc1> shouldn't "-vf fps=30" fix the speed? [20:03:17 CET] <bencc1> again, I'm looking for something similar for audio [20:03:24 CET] <kepstin> fps and speed are unrelated [20:03:53 CET] <bencc1> this doesn't force constant audio frames: "-acodec aac -b:a 96k -ar 48000" [20:03:55 CET] <kepstin> speed is whether the ffmpeg command is encoding video as fast as it's reading it, if it's lower than 1.0 that means it's falling behind realtime and stuff breaks [20:04:10 CET] <bencc1> ok [20:04:23 CET] <bencc1> I still think that it should be possible to force constant audio frames [20:04:39 CET] <bencc1> because it's working for video [20:05:03 CET] <kepstin> please look at the frame times in the nut file to narrow down where the issue is being introduced [20:06:24 CET] <kepstin> but if the problem's present there, it could possibly means that your system isn't reading audio fast enough, so it's missing bits. and there's no real way to fix that. [20:06:55 CET] <kepstin> please also pastebin the output, of the original capture ffmpeg command as well [20:07:17 CET] <bencc1> I will [20:07:31 CET] <bencc1> but I still looking for a way to force constant audio output [20:07:47 CET] <bencc1> it should be possible to duplicate and drop audio frames if ffmpeg doesn't keep up [20:07:57 CET] <bencc1> that's what I'm doing for video and it's working [20:10:52 CET] <kepstin> audio doesn't work like that :) [20:11:56 CET] <kepstin> that said, the resample filter *can* be configured to stretch audio to match timestamps, but iirc you'll get pitch shifting with that, and it won't fix dropouts. [20:12:14 CET] <kepstin> so if you're missing audio, it'll go from sounding bad to sounding bad differently. [20:12:37 CET] <kepstin> do you have an actual problem during playback, anyways? [20:13:51 CET] <bencc1> the video editing person complains that there is A/V sync issue when opening the mp4 in Adobe Premiere Pro [20:14:09 CET] <bencc1> there is no problem when playing the file in the browser or video players [20:22:23 CET] <bencc1> kepstin: I have two steps: 1) capture to nut 2) convert to mp4 without transcoding [20:22:40 CET] <bencc1> does it make sense to re-encode the audio to ensure constant frame rate? [20:23:01 CET] <kepstin> i have asked you for two things: output of the original capture command, and to look at the timestamps in the nut file [20:23:02 CET] <bencc1> maybe capture in higher quality to ensure I won't loss quality [20:23:13 CET] <bencc1> I will [20:23:20 CET] <kepstin> (actually, look at the packet timestamps too please) [20:23:40 CET] <bencc1> I need to run the capture again and it'll take 2 hours [20:23:58 CET] <kepstin> then, just run it for a few minutes to get a sample? [20:24:28 CET] <bencc1> ok [20:24:56 CET] <bencc1> trying [20:27:06 CET] <bencc1> kepstin: 1 minute will be enough? more? [20:27:58 CET] <kepstin> if your ffprobe output is right, it looks like the issue was happening even right at the start of the file, so it doesn't have to be too long. [20:28:59 CET] <bencc1> ok [20:40:55 CET] <asterismo_l> hi [20:41:23 CET] <asterismo_l> i'm having some problems on compiling ffmpeg without SSE and SSE4 cue support [20:41:49 CET] <asterismo_l> it seems that the VPS i'm running ffmpeg, does nat have this instruction sets enabled [20:42:16 CET] <bencc1> kepstin: 1 minute output: https://pastebin.com/raw/CJuJFc52 [20:42:28 CET] <asterismo_l> my question is, i only have to compile ffmpeg? or must i compile x264 too? [20:45:07 CET] <bencc1> kepstin: ffprobe -show_streams of the nut file: https://pastebin.com/raw/KQynq24j [20:46:48 CET] <bencc1> kepstin: ffprobe -show_frames of the nut file (first few frames) https://pastebin.com/raw/1aJvnZMn [20:46:58 CET] <bencc1> now audio packet duration seems constant [20:47:27 CET] <bencc1> I don't know if the problem is in the .mp4 file or because of server load or something else [20:48:53 CET] <kepstin> something's pretty seriously wrong there, hmm. you shouldn't be getting that "Queue input is backward in time" warning [20:49:05 CET] <kepstin> that means the timestamps pulseaudio is giving you are busted [20:50:53 CET] <bencc1> kepstin: is there something wrong with my capture command? [20:51:26 CET] <kepstin> i think the capture command is fine, although you might want to try without the -thread_queue_size options to make sure that's not the issue. [20:51:36 CET] <kepstin> it's possible you some sort of timing issue with the system clock or sound card [20:51:36 CET] <bencc1> ok [20:52:05 CET] <kepstin> (if this is a vm, virtual clocks behave... really poorly depending on configuration) [20:52:27 CET] <bencc1> this is a docker container on a google cloud vm [20:52:42 CET] <bencc1> is there anything I can do to improve the clock? [20:52:50 CET] <kepstin> ask google nicely [20:53:13 CET] <kepstin> (they'll ignore you, so that's a silly way of saying "no, get a better host") [20:53:33 CET] <bencc1> yes [20:53:47 CET] <bencc1> what about re-encoding the audio stream when converting .nut to .mp4? [20:53:51 CET] <bencc1> should it work? [20:54:51 CET] <asterismo_l> i'm using this tutorial for compiling x264. Is there any way to pass a disable sse3 and sse4 instruction flag? [20:54:55 CET] <asterismo_l> https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu [20:55:24 CET] <kepstin> bencc1: hmm... probably not, but you might as well try [20:55:42 CET] <kepstin> bencc1: I honestly still use dedicated hardware for all my realtime stuff to avoid problems like this :/ [20:56:18 CET] <kepstin> asterismo_l: x264 and ffmpeg automatically select which assembly to use depending on what the cpuid flags claim the cpu supports, you must have a really busted vm host :/ [20:57:51 CET] <asterismo_l> its like their hypervisor is not handling cpu well [20:57:56 CET] <bencc1> kepstin: I don't understand why re-encoding audio won't work [20:58:49 CET] <asterismo_l> https://paste.debian.net/1122749/ [20:58:56 CET] <kepstin> bencc1: if everything's working correctly, re-encoding will preserve timestamps, so the audio will have the same issue where audio timestamps are going up faster than can be accounted for by the number of samples. [20:59:05 CET] <bencc1> kepstin: that's interesting: "ffprobe -show_frames" for the nut file show constant audio frame duration [20:59:18 CET] <bencc1> but audio duration for the mp4 varies [20:59:24 CET] <kepstin> bencc1: the difference between pts of successive frames is the important bit [20:59:27 CET] <bencc1> any idea why? [20:59:28 CET] <kepstin> bencc1: ignore the durations [20:59:39 CET] <bencc1> ok [21:00:20 CET] <kepstin> asterismo_l: honestly, i'd recommend complaining to your vm host so you can get a refund so you can switch to a better host which doesn't have broken cpus :/ [21:01:11 CET] <kepstin> asterismo_l: you can check the x264 and ffmpeg configure scripts, there might be options to disable assembly for specific instruction sets [21:01:34 CET] <kepstin> but the encoder does run a fair bit faster if you get a cpu that supports those, so... :/ [21:01:38 CET] <bencc1> kepstin: your suggestion is to check the machine/vm/container clock? [21:01:53 CET] <bencc1> kepstin: no post-processing or command param to try? [21:02:02 CET] <kepstin> bencc1: i assume there's no physical sound card? [21:02:24 CET] <bencc1> no [21:02:25 CET] <kepstin> bencc1: any command that would "fix" the audio timestamps would likely cause a/v desync [21:02:32 CET] <bencc1> tried without "-thread_queue_size". same result [21:02:59 CET] <kepstin> bencc1: then check the pulseaudio log output, i'd guess it's running on high-res timers, and is probably complaining about the problem :/ [21:03:14 CET] <bencc1> I thought that transcoding the audio means aac->raw->aac so I can "fix" it [21:03:26 CET] <kepstin> bencc1: that would "fix" it, but would also cause a/v desync [21:03:29 CET] <bencc1> where is the pulseaudio log? [21:03:40 CET] <kepstin> depends how you started pulseaudio [21:03:46 CET] <bencc1> ubuntu [21:03:52 CET] <bencc1> sorry, debian [21:03:59 CET] <bencc1> started with: "pulseaudio --start" [21:04:32 CET] <kepstin> hmm, this is a docker container, you said? it wouldn't have a user journal then [21:04:42 CET] <kepstin> i honestly don't know what it would do [21:05:26 CET] <bencc1> I'll try to add log target to the command [21:05:27 CET] <bencc1> testing [21:06:30 CET] <kepstin> bencc1: what are you capturing anyways? are you sure it's not possible to use it as an input directly, so you don't have to do realtime capture of application output? :/ [21:07:35 CET] <bencc1> I'm capturing chromium browser running in xvfb virtual display [21:08:36 CET] <kepstin> if you weren't in docker, i would suggest trying aloop, since it runs in the kernel, it might not have as bad timing issues [21:08:58 CET] <kepstin> note that pulseaudio likes to run with realtime scheduling mode, which probably also doesn't work in docker [21:10:14 CET] <kepstin> i used to have an application where i was using a browser in xvfb to render a video, and i ended up rewriting it to be a python script that drew the thing directly with pycairo and sent raw frames to ffmpeg to encode [21:10:20 CET] <kepstin> so... yeah. [21:10:31 CET] <kepstin> (runs faster than realtime now) [21:10:39 CET] <bencc1> cool [21:11:17 CET] <bencc1> does it run faster than the speed of light? [21:12:07 CET] <kepstin> it was a batch processing thing, not live input, so having the browser render it realtime was the limiting factor on speed. [21:12:32 CET] <bencc1> I tried this: 'pulseaudio --start > ./pulse.log 2>&1 [21:12:41 CET] <bencc1> but log file is empty [21:12:48 CET] <bencc1> I'll try raising the log level [21:14:06 CET] <kepstin> you'd have to run pulseaudio in the foreground (without --daemonize - note that --start implies --daemonize) for it to log to stderr. [21:14:30 CET] <kepstin> you can use --log-target to tell it to log to a file instead tho [21:14:56 CET] <kepstin> (i found that out by reading the man page just now) [21:14:58 CET] <bencc1> trying [21:19:18 CET] <another> asterismo_l: update your ffmpeg [21:19:36 CET] <another> iirc this is a known bug which has been fixed in recent versions [21:19:42 CET] <asterismo_l> another, i'm compiling it from source [21:20:05 CET] <another> good [21:20:16 CET] <another> but you don't need to disable sse* [21:20:16 CET] <asterismo_l> wget -O ffmpeg-snapshot.tar.bz2 https://ffmpeg.org/releases/ffmpeg-snapshot.tar.bz2 [21:33:41 CET] <bencc1> kepstin: pulseaudio log: https://pastebin.com/JvAiqzJd [21:36:27 CET] <kepstin> well, you should probably be recording at 44100Hz instead of 48000Hz to avoid an extra resample (and extra cpu usage) [21:36:42 CET] <kepstin> but yeah, pulseaudio is definitely complaining about a bunch of buffer underruns there [21:37:56 CET] <bencc1> where do you see buffer issues in the log? [21:38:42 CET] <kepstin> there's a bunch of "Requesting rewind due to end of underrun" [21:38:59 CET] <kepstin> and a rewind sounds like something that would generate broken back in time timestamps [21:39:26 CET] <bencc1> ok [21:40:35 CET] <bencc1> anything I can do about it? [21:40:40 CET] <bencc1> beside using 44100Hz [21:41:12 CET] <kepstin> bencc1: ask the pulseaudio people for help, i guess? [21:41:33 CET] <kepstin> trying to get it to run with realtime scheduling might make a difference, but i have no idea how you'd do that [21:41:57 CET] <bencc1> thanks [21:57:47 CET] <BtbN> Hm, I want to concat and stream a ton of mostly unrelated videos out via rtmp. First step is that I transcode them all using the same settings. Step two I create a m3u8 playlist with discontinuity tags with all of them in, and play it with -re. [21:58:23 CET] <BtbN> Result: On the boundaries of the videos, it freezes for a few seconds, then the audio audibly fast-forwards, playback resumes, but a/v is horribly off-sync?! [22:19:01 CET] <kepstin> BtbN: you might be better off using the ffmpeg's concat demuxer/playlist format [22:21:53 CET] <BtbN> kepstin, doesn't make a difference [22:22:04 CET] <BtbN> the files can be in vastly different resolutions/framerates/... [22:22:16 CET] <BtbN> that's why I opted for a local HLS playlist, so I can at least signal that [22:22:31 CET] <BtbN> was hoping throwing a scale and fps filter in there would just transform everything. But it doesn't. [22:22:50 CET] <kepstin> hmm, the issue might be due to how ffmpeg opens each file as it comes up in the playlist, so there's a delay as it does that [22:23:09 CET] <BtbN> A few seconds of being "stuck" shouldn't be an issue [22:24:04 CET] <kepstin> you could try doing something like having the ffmpeg reading the playlist output to something like nut in a pipe to a second ffmpeg that does the rtmp stream (the latter would have -re, the first would not) [22:24:34 CET] <BtbN> This would be a bit messy to setup, since it's not a shell calling ffmpeg here. [22:26:53 CET] <BtbN> I also tried just two-step concatenating everything into a temporary file first, and then streaming that. But it shows the same behaviour. [22:27:27 CET] <kepstin> hmm, i'd really expect that to work. [22:27:45 CET] <kepstin> must be something weird going on with timestamps then [22:27:48 CET] <BtbN> I feel like Twitch horrily messed up the timestamps in those VODs [22:27:54 CET] <BtbN> (They are downloaded highlights) [22:28:30 CET] <kepstin> worst case if you do them as separate ffmpeg inputs and use the concat muxer, that should apply all the timestamp cleanups and give you something reasonable [22:28:40 CET] <kepstin> sorry, concat filter* [22:29:15 CET] <BtbN> Yeah, that's my next idea, but will be a behemoth of a commandline, since I'll need to do it for audio and video, and lists can have a couple dozen videos [22:29:56 CET] <BtbN> And it'll probably explode if there's the slightest discrepancy between audio and video length in one of the inputs? [22:30:20 CET] <kepstin> concat filter corrects mismatched audio/video lengths in the inputs [22:30:33 CET] <BtbN> But isn't it only a video filter? [22:30:42 CET] <kepstin> no, it's an a/v filter [22:30:51 CET] <BtbN> ah, nvm that then [22:31:10 CET] <kepstin> you can use it on only video or only audio if you want, of course [22:31:24 CET] <kepstin> but it can also work on arbitrary combinations of audio+video tracks. [22:31:43 CET] <BtbN> Looking at the downloaded video file from one of the Highlights... even that is already messed up [22:31:53 CET] <BtbN> so Twitch is definitely doing shit with their timestamps [22:36:14 CET] <BtbN> This Video, for example: https://www.twitch.tv/videos/456369484 [22:36:21 CET] <BtbN> Downloading it via ffmpeg -i https://vod-metro.twitch.tv/d3ecb02f27b35de68d4e_esamarathon_49108928056_50… -vf scale=1920:1080,fps=60 -c:v libx264 -crf 26 -preset veryfast -c:a aac -ar 48000 -b:a 192k test.ts [22:36:43 CET] <BtbN> generates a TON of [mpegts @ 0x55d8862f0300] Non-monotonous DTS in output stream 0:1; previous: 345547, current: -1527330; changing to 345548. This may result in incorrect timestamps in the output file. [22:37:02 CET] <BtbN> So yeah, they are returning garbage timestamps [22:43:39 CET] <BtbN> But somehow only for the first ~40 seconds of that video [22:48:44 CET] <BtbN> Maybe I need https://github.com/FFmpeg/FFmpeg/commit/d6ac6650b911f0957e69545d7fc25be6b77… [23:12:50 CET] <BtbN> Yeah, that seems to work _a lot_ better. [23:29:20 CET] <vlt> Hello. What does filter "loudnorm" do? Just change volume depending on RMS peak or similar or will it compress? [23:32:04 CET] <CoreX> read the documentation https://ffmpeg.org/ffmpeg-filters.html#loudnorm [23:36:52 CET] <vlt> CoreX: Thank you! [00:00:00 CET] --- Fri Dec 27 2019

1 0