significant lag in vrecord -a

[metacynicv2]

There is a 10s delay between the image in the source and the image on the vrecord audio passthrough window (vrecord -a). This makes it difficult to calibrate and make adjustments.

[metacynicv2]

It is also crashing after several minutes

[privatezero]

I want to fix this!!! But currently this seems to be a limitation of the ffplay filterchains vrecord is using. They just can't keep up once audio information is introduced. For now I would suggest video calibration using the standard passthrough mode, and then using the -a function just for things like confirming your audio signal is coming through to the appropriate channels.

[dericed]

cc'ing @richardpl for any advice on getting streams=dv+da[v][s];[s]asplit=2[z][out1];[z]channelsplit=channel_layout=4.0[s1][s2][s3][s4];[s1][s2][s3][s4]amerge=inputs=4,aformat=channel_layouts=quad[x];[x]showvolume=t=0:h=17:w=200[xx];[v]split=5[a][b][c][d][e];[b]field=top[b1];[c]field=bottom[c1];[b1]${WAVEFORM_FILTER}[b2];[c1]${WAVEFORM_FILTER}[c2];[a][b2][c2]vstack=inputs=3,format=yuv422p[abc1];[d]${VECTORSCOPE_FILTER}[d1];[e]signalstats=out=brng,scale=512:ih[e1];[e1][d1]vstack[de1];[abc1][de1]hstack[abcde1];[abcde1][xx]overlay=10:10[out0] to keep up with a real time standard definition NTSC source.

[richardpl]

I doubt this can be fixed in ffplay, time to switch to something better like mpv?

[richardpl]

Is signalstats metadata stuff mandatory? It appears to be major CPU consumer.

[dericed]

hmmmm, i suppose we could consider audio or signalstats. But when calibrating the video decks I find the signalstats/drawtext combination very helpful.

[richardpl]

I see, what's CPU usage without and with showvolume?

[richardpl]

Either there is no enough CPU power or buffer management in lavfi is bad. Note that even if your CPU have multiple cores lavfi may not use all of them. I could reproduce with mpv somehow. But verbose ffplay output would be helpful. I don't have ffplay right now installed....

[privatezero]

I have been experimenting today with using mpv instead of ffplay for vrecord -a, so far it is looking promising. Seems like it is keeping up with the stream for both a file input source as well as a blackmagic input. It doesn't like the signal stats (visual + numerical) mode yet, however.

Here is the filter chain I have been testing so far: [aid1]asplit=2[z][ao],[z]showvolume=t=0:h=17:w=200[xx],[vid1]split=5[a][b][c][d][e],[b]field=top[b1],[c]field=bottom[c1],[b1]${WAVEFORM_FILTER}[b2],[c1]${WAVEFORM_FILTER}[c2],[a][b2][c2]vstack=inputs=3,format=yuv422p[abc1],[d]${VECTORSCOPE_FILTER}[d1],[e]signalstats=out=brng,scale=512:ih[e1],[e1][d1]vstack[de1],[abc1][de1]hstack[abcde1],[abcde1][xx]overlay=10:10[vo]"

[richardpl]

Too much stacking invocations for my taste, I'm wondering about adding generic stacker which would accepts multiple inputs at once and just copy them to random places in new frame. As a result it would use less memcpy calls.

[privatezero]

Merged this. Still doesn't support numeric information in audiopassthrough, but otherwise everything else seems to be keeping up on this end! Hopefully this helps Kelly!

[retokromer]

What is the status of this issue?

[privatezero]

I assume that the switch to mpv resolved this? It was working fine on our end after the switch.

[retokromer]

Thank you! That was my impression too. Then I close this one.