Re: [codec] Opus comparison test plan version 2

["Christian Hoene" <[e-mail]>]

Hi,

 

As for 24, 32, 48 kbit/s for mono and 32, 48, 64 and 96 kbit/s for stereo we
can assume that least one out of the two codecs will work. Thus, we keep
those parameter setting in.

 

With best regards,

 

Christian

 

 

 

Von: [e-mail] [mailto:[e-mail]] Im Auftrag von
Alfons Martin
Gesendet: Mittwoch, 17. April 2013 15:50
An: [e-mail]; [e-mail]
Cc: Christian Hoene; [e-mail]
Betreff: [codec] Opus comparison test plan version 2

 

 

AMR-WB

 

Based on the feedback we changed the document.

1)      No input bandwidth filtering for speech at all

2)      "Make sure to tell the Opus encoder what the percentage of loss is
so it can optimize the encoding for it." - Ok, but we have to consider both
cases.

3)      "For AMR-WB the 8.85 mode is "intended to be used only temporarily
during severe radio channel conditions or during network congestion". Right,
better we use AMR-WB 12.65 with DTX in this case.

 

We did not consider the following suggestion

1)      "Opus should use VBR" - For direct comparison, we have to choose
same mode as the other codec. For example, AMR-WB does not support VBR.

2)      "If you're going to test packet loss, there should be at least one
test with FEC -- probably the 23.85 one". As for now, we are not sure
whether the Opus FEC implementation is bug free (see FEC question email).

 

AAC-eLD

Based on the feedback we changed the document.

1)      We will use SBR

2)      "Mono and stereo should not be mixed in one experiment to avoid the
influence of this factor on the assessment of the coding quality." - agreed.

3)      24, 32, 48 kbit/s for mono at 32 kHz sample rate CHECK 

4)      For stereo 32, 48, 64 and 96 kbit/s could be used. CHECK

 

We did not consider the following suggestion

1)       

 

 

 

General

Based on the feedback we changed the document.

1)      We will use loss rates 0, 1, 3, 6%

 

We did not consider the following suggestion

1)      We will keep the MNRU 16 anchor (MOS 2.2) because some of the
samples will be worse than LP 3.5 but better than MOS 2.2. We skip LP 7 as
it is optional.

2)      We stick to MUSHRA as it is the only tests that covers a wide range
of degradations.

 

Thank you very much for your timely feedback.


Alfons, Christian