TEST RELATED TO THIS POST


• Nero AAC 3.0.0.4 VBR ::internet :: ‘fast mode’ [for 'fast' mode, see this test].
• Vorbis 1.1 –q 4
• Vorbis Quantum Knot 2005.02.28 based on 1.1, -q4



Edit: comments in blue or yellow are only summuries (and are therefore simplistic). I didn't mention usual vorbis issues, like HF boost or coarseness.

ABC logs are available here.


Samples :
not available anymore



SHORT (and general) CONCLUSIONS

• Nero AAC is less agressive on microdetails samples (no or very limited blurring), but introduces annoying distortions with most microattacks samples. The encoder suffers sometimes from excessive lowpass.

• Vorbis smooths too much small musical asperities (microdetails) at this bitrate. But with stronger details (microattacks), artefacts are less annoying to my ears than with Nero AAC: vorbis adds noise, and only noise, whereas Nero AAC adds some irritating distortions.

• Vorbis QK28 apparently doesn't help at all with this kind of samples.

Thanks for the tests. It would be interesting to do a mini-test (on 1 or 2 samples where my encoder sounded no different to 1.1) using the 20050227 version, rather than 20050228. I have a feeling that my quick fix for excessive bitrates on low frequency tonals has broken some things.

I could do it. Next week

BTW, (if this has'n been discussed before):

Wouldn't it make more sense to rate a sample dependent on the bitrate?

Eg using rating = subjective_rating * reference_bitrate/actual_bitrate ?

Of course this alone is not enough, as in VBR one is interested in getting constant quality, so the table should give both, subjective_rating and rating considering bitrate.

It has been discuted many times.
It's not pertinent in my opinion. First, because the average bitrate of a short sample doesn't correspond to the average bitrate of the track, which is also often different from the average bitrate "in general" of the preset/setting.
If you try some of these samples with mpc --quality 4, bitrate could reach 390 kbps ! But the average bitrate for the complete track is close to 170 kbps, and average bitrate for the album encoded with --quality 4 drops to 130 kbps... Users generally listen to full tracks or album, and not to short samples. Therefore, the only pertinent bitrate value is the bitrate of the title/song.

Second reason: quality is not linear. Lowering the notation by 50% because bitrate is higher to 50% from the target lead to a non-pertinent notation.

Ok, thanks for the detailed and sense-making explanation.

@guruboolez:
Can you also test with Helix DNA producer AAC codec?

I don't have any Helix softwares on my computer. Did the AAC encoder provided by Coding Technologies changed since last year? Is it still CBR only? Is the lowpass at CBR128 still fixed to 15 KHz?

Here's the Real video 9/10 informations
The change log seems only have HE-AAC info...

@guruboolez:
you were right, the Helix DNA producer aac 128kbps encoded audio still at 15KHz lowpass(check with RMAA, but i'm not sure this is the right method...), and still CBR only... So, forget about it.

BTW i'm interest with the Helix DNA producer HE-AAC codec, and their encoded audio @ 96kbps seems have higher lowpass(~20KHz) than it own LC-AAC codec,
can you please make Helix DNA producer HE-AAC vs. Nero HE-AAC listening test?

My time is fortunately limited And SBR encodings at 96kbps aren't worth in my opinion. Therefore, I probably won't test it in the next time.

If you are free please make this test, otherwise please forget it
SBR only can go up to 96kbps only...(for Helix HE-AAC)
Thanks for the replies

Why don't you conduct a listening test yourself? Its not like it would be that difficult at those bitrates.

Just looking at the lowpass is silly. At 96kbps LC AAC has a good chance to be better than HE-AAC, even if it has a lower lowpass.

Why? Because i don't have good hearing, don't have killer sample(?),
don't know how to use ABX(inside fb2k), don't have enough experience to do a listening test

Experience can only be acquired by practive

Ok, i'll try my best...

I was wondering... was this ever tested? It would be nice to see this improvement merged in aoTuVb4 if it really helps :)

You know what: I forgot to do my homeworks


Sorry...