Re: [codec] A concrete proposal for requirements and testing

[Paul Coverdale <[e-mail]>]

Hi Marc,

I think that the issues of performance requirements and encumbered
technologies are separate. In fact, the WG charter states "The working group
cannot explicitly rule out the possibility of adopting encumbered
technologies". At the end of the day, people will make their own trade-off
between Opus performance and whether it is encumbered or not. So the
requirements need to include some reasonable performance targets, and AMR-NB
and AMR-WB provide those.

In any case, the test plans that I sent out earlier are examples, and need
to be viewed in relation to the terms of reference that were agreed for
G.729EV. The intention at the time was to provide scalable wideband on top
of G.729, so for NB the main comparison was with G.729A. For WB, the chosen
comparison was primarily G.722, with some G.722.2.

Regards,

...Paul

-----Original Message-----
From: Jean-Marc Valin [mailto:[e-mail]] 
Sent: Wednesday, April 06, 2011 9:37 PM
To: Paul Coverdale
Cc: 'Anisse Taleb'; [e-mail]
Subject: Re: [codec] A concrete proposal for requirements and testing


I strongly disagree with including AMR-NB and AMR-WB as strict 
requirements for several reasons. Those codecs are not credible 
alternatives to Opus within the charter of this WG, to "try to avoid 
encumbered technologies."

Also, I note from the ITU test plan for G.729.1 (then G.729EV) you sent 
that it does not even mention AMR-NB. As for AMR-WB (G.722.2), it is 
only included at a bit-rate of 8.85 kb/s even though the lowest G.729.1 
bit-rate is (AFAIK) 14 kb/s. Looking at this paper from Nokia: 
http://research.nokia.com/files/public/%5B11%5D_ICASSP2010_Voice%20Quality%2
0Evaluation%20of%20Various%20Codecs.pdf

we see that G.729.1 is clearly worse than AMR-WB. Despite that, it seems 
like the ITU-T has seen the value of this codec, which supports both 
narrowband and wideband at multiple rates. The codec we're designing 
here is meant to do not only narrowband and wideband, but also 
super-wideband and fullband, for both speech and music and for mono and 
stereo. This is something which -- to the best of my knowledge -- has 
not been done before for an interactive codec. Requiring Opus to also 
beat all codecs out there in all circumstances is thus totally 
unreasonable (despite the fact that the tests we did so far indicate we 
may be close to that).

Cheers,

	Jean-Marc