Re: [codec] Opus comparision test plan

[Jean-Marc Valin <[e-mail]>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 04/16/2013 10:04 PM, [e-mail] wrote:
> For the test it should be ensured that the used codec
> implementation fulfills  high quality requirements. I don't know
> the AAC-ELD fdk-aac library.  The link in your document did not
> work for me. It does not make sense to test an AAC-ELD encoder that
> does not provide state-of-the-art encoding quality.

Well, last I heard when Jan Skoglund was organizing a music, there was
just no way for him to obtain any LD or ELD implementation (IIRC FhG
refused to send him one). That being said, I've no idea what the
quality of tdk-aac looks like.

> Concerning SBR: AFAIK know AAC-ELD standard does not have SBR
> specified as a tool.

Actually SBR is the main improvement that ELD has over LD.

> I think, 20 % Frame erasure rate makes no sense as it is
> unrealistic. I would prefer frame erase rates like 1%, 3% or  6%.
> These rates should be used both in speech and music experiments.

I tend to agree on that one.

> For music testing I would recommend  the bitrates 24, 32, 48
> kbit/s for mono at 32 kHz sample rate.

This is way too low. You end up in a zone where neither codec will
provide acceptable quality music and where the result is highly
sensitive on the exact way in which the encoder implementation decided
to mitigate the "falling apart" case.

> For stereo 32, 48, 64 and 96 kbit/s could be used.

48 kb/s stereo is pushing it. 32 kb/s is just ridiculously low for
stereo and requires parametric stereo, which neither AAC-ELD nor Opus
has (not that PS sounds particularly good in the first place).

> I think these working points would be used most likely in
> applications where low delay is a major requirement.

For low-delay applications, it's useless to try saving 8 kb/s when
you're spending 32+ kb/s transmitting headers.

> It is more important to test a higher number of contents than to
> tests the very high bitrates.

Totally agreed.

> What is a MOS 4 anchor in the context of a MUSHRA test?

Not sure what you mean by "MOS 4 anchor", but MUSHRA specifies that
you must have a 3.5 kHz low-pass anchor and optionally anchors with a
low-pass at 7 or 10 kHz (in this case 7 is more appropriate IMO).

Cheers,

	Jean-Marc
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.13 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJRbgb2AAoJEJ6/8sItn9q9VewIAL3Gbp4cdXibNs+4SM5c65sg
0ESk7gbhwc8wxdzTS0lslxwhCvdOY2ELZ8HOQBeYnISt45JyWJ7fuRrgBI0pnfPh
rp6eVLOB8R9YY80+rRL/t9F9lmC15D//TV63C19Nn2zXrdfQXUKICdG+s5Q6ez9r
9bdC+SpXwxpU/Gvt0pXPWf8L8XdNAxDGEOvy71FLsLbOL5G4C7zeawGpMF7VFSTD
el1HmbF7jkh66YHSQ/KNomG53gw+NXNMRBSgWky8lupPX9ybj/MJvGuSOLNCQ1je
iLyeejiKW5sWlJ1vXxQbE+uHUTgTWOhfdn2O2nrIAAVzBjNrP1RLtOgnp9YwAos=
=6qMu
-----END PGP SIGNATURE-----