[opus] [codec] Some listening test results

[Timothy B. Terriberry]

Ron wrote:
> I'm curious about that because you note that the Opus bitstream
> wasn't designed to be inherently robust against a bit error at
> any point in the packet (and some places will obviously effect
> it worse than others) - however it's not quite true that it
> 'cannot tolerate them'.

It's not quite true to say that Opus was not designed to be robust to 
them, either. We use "raw bits" (RFC 6716 Section 4.1.4) where possible 
precisely because a bit error in one of those will not desync the 
bitstream. We also designed things so that the bits most sensitive to 
errors were at the beginning of the packet (see slide 46 of 
<http://www.celt-codec.org/presentations/misc/lca-celt.pdf> for some 
early measurements on the CELT layer from my 2009 LCA talk). This takes 
advantage of the unequal error protection afforded by Trellis Coded 
Modulation (TCM), and in other modulation schemes a SECDED code on the 
first 64 bits or so of the packet should substantially cure the bulk of 
bit-error impairment.

As Ron notes, the choice of modulation scheme or error protection scheme 
is best dealt with at a layer above the codec itself. We just tried to 
make it easy to do so. I guess you can argue whether or not that's 
"inherent".