[opus] over/underestimating expected packet loss

Jean-Marc Valin jmvalin at jmvalin.ca
Sat Jul 6 21:13:53 PDT 2013

On 07/07/2013 12:01 AM, Marc Lindahl wrote:
> I'm curious what happens if you feed an inaccurate value to 'expected loss'?
> Channel loss rate isn't always easy to measure, esp. in real time.
> I'm trying to figure out, ultimately, is it a better strategy to put the worse case expected loss in there, or more like the best case?
> (dealing with audio/full bandwidth here)

It's not that big a deal. You can always feed a measured average, a
guess, ... but probably not a worst-case. Basically, it's a way to trade
loss-free quality for robustness. Using a high value means that the
audio will be slightly worse when there's no loss, but it will recover
more quickly in case of loss. But don't worry, nothing terribly wrong
will happen if you get it wrong. So don't spend too much time tweaking this.



More information about the opus mailing list