[tialaramex]

Jitsi has audio problems for several people in my peer group. I think from other audio work that it may be that they're using too small frame sizes, although it's possible that something else in their audio pipeline is busted.

Audio capture APIs often suggest it may be possible to use very small frame sizes, which naturally promise much improved latency. Going from 100ms of audio latency to 20ms is great so surely going from 20ms to 5ms is even better right? Well, the hardware underneath that API may not be able to deliver, at least it may not be able to deliver consistently. If your 5ms buffer isn't filled on time, what do you send? A partially filled buffer? Silence? The last 5ms of filled buffer again? All bad answers.

Tool A with 40ms of latency may feel imperceptibly worse than Tool B with 30ms of latency. But Tool C with 10ms of latency but frequent "drain piping" as audio frames are garbled or undelivered is clearly much worse than either.