Finally read this. They consider a half second of delay as "good".
One of these days I'd like (someone) to tackle these problems, using a
2.7ms interval opus voip flow as a clock to regulate the size of the
video stream, in an fq_codeled environment, striving for a
buffer-related video delay of under 20ms.
I also miss scan-line video capture devices. 16ms minimum lost in the
usb side already.
Post by Toke Høiland-Jørgensen
From the good people over at Stanford: https://snr.stanford.edu/salsify/
Bloat mailing list
CEO, TekLibre, LLC