Am Donnerstag, 9. Februar 2017 schrieb christina zou : > I've made some progress thanks to advice from the IRC channel. I increased > my pipe max buffer size as high as I could: > > sysctl fs.pipe-max-size=1048576 > > I also lowered my video resolution and bitrate. > > Now, with the below command I've managed 3 minutes of fully synced > audio/video livestream to Youtube from my Pi Zero. At the 3 minute mark, > there's a slight desync, and the audio goes about 0.5s ahead of the video. > > That's my next question. I don't mind if there's a stutter every 3 minutes, > but I need the A/V sync to recover after that. So if I lose a few audio > frames, I need the video to catch up (or audio to slow down). Any > suggestions about how to do this (either ffmpeg param, or script)? > > Here's my current command: > > sysctl fs.pipe-max-size=1048576 > > *sudo rm temp_audio.v* > > *sudo rm temp_video.h264* > > *mkfifo temp_audio.v* > > *mkfifo temp_video.h264* > > *arecord -Dmic_sv -c2 -r48000 -fS32_LE -twav temp_audio.v & \* > > *raspivid *-w *640* -h *480** -fps 10 -v -b 1000000 -o temp_video.h264 -t 0 > & \* > > ~/special/ffmpeg/ffmpeg* \* > > * -framerate 10 \* > > * -i temp_video.h264 \* > > * -i temp_audio.v \* > > * -ab 24k \* > > * -c:v copy \* > > * -c:a aac \* > > * -report \* > > * -f flv **rtmp://209.85.230.23/live2/KEY* > > On Wed, Feb 8, 2017 at 3:50 PM, christina zou <[email protected] > <javascript:;>> > wrote: > > > Carl, > > > > I spent a lot of time trying to use ffmpeg alone (alsa/v4l2), but it > > produces extreme stuttering on the audio track. Using my approach of > > writing to named pipes and sending those into ffmpeg, I add a bit of > > latency but ensure that there's no loss of frames. This was the approach > > suggested to me by the ffmpeg IRC channel. I'm on a Pi Zero (single > core), > > by the way. > > > > Is there any way to align the two input streams' timestamps, if they are > > both being read from named pipes? > > > > Here is my full console output: http://pastebin.com/phmiphaL > > > > Thanks, > > Christina > > > > > > > > On Wed, Feb 8, 2017 at 3:02 PM, Carl Eugen Hoyos <[email protected] > <javascript:;>> > > wrote: > > > >> 2017-02-08 22:57 GMT+01:00 christina zou <[email protected] > <javascript:;>>: > >> > >> > *arecord -Dmic_sv -c2 -r48000 -fS32_LE -twav temp_audio.v & \* > >> > >> Why? > >> (See below) > >> > >> > *raspivid -fps 10 -v -b 3000000 -o temp_video.h264 -t 0 & \* > >> > >> Is this not possible with ffmpeg alone? > >> > >> > ~/special/ffmpeg/ffmpeg* \* > >> > * -framerate 10 \* > >> > * -re \* > >> > * -i temp_video.h264 \* > >> > * -i temp_audio.v \* > >> > >> (Complete, uncut console output missing.) > >> How are these two input streams supposed to be synced? > >> > >> Normally, you would use (for example) alsa and v4l2 input > >> and hope that the two drivers both provide wallclock timestamps. > >> > >> Carl Eugen >
My suspicion is the transcoder the stream is being fed to doesn't cope well with the framerate. In my experience it's most stable when using either 25 or 30fps with a keyframe interval of 50 or 60 frames maximum. Also best to use standard bitrates, especially with audio, to avoid drifting. Erik _______________________________________________ ffmpeg-user mailing list [email protected] http://ffmpeg.org/mailman/listinfo/ffmpeg-user To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
