1y ago

Ffmpeg makes choppy videos when recording audio

Well, my issue is simple to explain, but I tried everything to solve it without success.

I like recording gameplays, just for fun. The game I currently make recordings for is not resource heavy (it's Doom 2 through GZDoom). I'm using Lubuntu 22.04.

So, here is the command I've just used:

ffmpeg -f x11grab -draw_mouse 0 -framerate 30 -probesize 42M -video_size 1920x1080 -i :0 -f pulse -i default -c:v libvpx-vp9 -r 30 -g 90 -quality realtime -frame-parallel 1 -qmin 4 -qmax 48 -b:v 4500k -c:a libvorbis prueba.mkv

The issue is, if I record without including the audio, then the video plays smoothly after being recorded. But if I record including the audio, no matter what parameters, codecs, bitrate or anything I use, then the video is barely watchable, with lots of choppiness, and even with delay between audio and video, apparently.

SimpleScreenRecorder works fine, but sometimes there are few frames dropped and I don't like it much. OBS is very heavy for my hardware (Mac Mini 2016, Intel Core i5 1.4 GHz 4260U, Intel HD Graphics 4000, 4 GB RAM).

I also tried VAAPI, with same results as above.

And the thing is that I remember recording some years ago, with no issues at all.

I don't know what to do. Any tip or solution?

8 comments

Does top/htop/whatever show a large difference between encoding with and without audio? Do e.g. the pulse processes hog resources for some reason?
Do you get smooth recording if you reduce the screen resolution and do something that uses minimal resources (e.g., play Tetris at 640x480)?
- Depending exclusively on the codecs I use, htop shows a CPU usage of 90-100% or 10-20%. So, I use those that don't require a huge CPU usage. Memory and Swap are not affected, as far as I know. But despite those circumstances, the outcome is the same. Recording with no sound = smoothness, 60 fps, recording with sound = choppiness, 10 fp5s.
  Something I also noticed is that it also happens before I open Doom 2, so it doesn't matter whether I'm playing or not. Even with just the terminal and the file manager open.
  
  Crazy idea, but what happens if you record audio and video with different ffmpeg processes? Just for debugging this might be useful to try. Perhaps there's something funny about the timestamps you're receiving from Pulse that causes weird behavior (or something along those lines).
  Also, you mentioned this works better with other software --- any idea if those settings are drastically different than your ffmpeg settings?
  I have no idea what I'm talking about though, just throwing out ideas :)
Here are my 2 cents. Try to use some codec with hardware acceleration like h264_qsv but make sure that your version supports it, otherwise you will need to re-compile it. Try to simplify your ffmpeg command and update your ffmpeg version if it is not the latest one. Example command:

ffmpeg -f x11grab -draw_mouse 0 -framerate 30 -probesize 42M -video_size 1920x1080 -i :0 -f pulse -i default -c:v h264_qsv -preset:v ultrafast -r 30 -g 90 -b:v 4500k -c:a aac prueba.mkv
In this command:
-c:v h264_qsv specifies the H.264 video codec with Intel Quick Sync Video hardware acceleration.
-preset:v ultrafast - sets the video preset to ultrafast, meaning that ffmpeg will try to lower the quality of encoding and should potentially decrease your CPU utilisation. In this case I would also recommend you to increase the -b:v to 6-8-10M to compensate for the quality loss and if needed you can always offline transcode it to lower the size of your recording
-c:a aac specifies the AAC audio codec for audio encoding.Other parameters such as framerate, video size, input sources, and output file name remain the same as in your original command.
Make sure your version of ffmpeg is compiled with support for Intel Quick Sync Video. You can check this by running ffmpeg -encoders and looking for the h264_qsv encoder in the list. If it's available, you should be able to use hardware encoding with Intel HD 4000.
Another idea is to try to run video grab only and then audio grab only and check the CPU utilisation. You can try to change the

-f pulse -i default
and define the input manually, and not leaving it to default. Also make sure you are not using ALSA or pipewire, as the audio grab syntax is slightly different from what you posted

8 comments