neroAacEnc.exe and 6ch encoding efficiency ... ? Options

[Sagittaire]

I have little ac3 6ch 448 Kbps. I make these convertions:
file wav 6 chanels 6ch.wav
file wav 2 chanels 2ch.wav (default matrix 6ch->2ch conversion for BS)
six files wav FR.wav, FL.wav, C.wav, LFE.wav, SR.wav and SL.wav

1) Convertion with neroAacEnc.exe
I use this command line for all wav file
neroAacEnc.exe -q 0.3 -lc -if input.wav -of output.mp4

and I find these size for each mp4 files
2ch done 2497 Ko and 114 Kbps
6h done 8764 Ko
FL done 1574 Ko
FR done 1569 Ko
C done 1567 Ko
LFE done 725 Ko
SL done 1481 Ko
SR done 1533 Ko

well it's really curious:
- same quality level for individual chanel done less bitrate than 6ch conversion (8.32 Mo vs 8.55 Mo)
- for same quality level bitrate for 6ch and 2ch sould be simple scalling something like between x2.0-x2.5 ... here it's 3.5


2) Convertion with FAAC.exe
I use this command line for all wav file
faac.exe input.wav -o output.mp4 --tns -q 100

and I find these size for each mp4 files
2ch done 2552 Ko and 116 Kbps
6h done 6559 Ko
FL done 1212 Ko
FR done 1211 Ko
C done 844 Ko
LFE done 2497 Ko
SL done 1233 Ko
SR done 1240 Ko

well here it's normal:
- same quality level for individual chanel done more bitrate than 6ch conversion (7.00 Mo vs 6.40 Mo)
- for same quality level bitrate for 6ch and 2ch sould be simple scaling something like between x2.0-x2.5 ... here it's 2.57

If quality level is the same for each encoding (1.0, 2.0, 5.1 ... etc) then bitrate for q 0.3 6ch file should be in 5000-6500 Ko interal. Chanel coupling seem work for 2.0 but not for 5.1. Bug for multichanel or different quality level ... ???

This post has been edited by Sagittaire: May 6 2006, 02:31

[kwanbis]

http://www.hydrogenaudio.org/forums/index....ndpost&p=389146

[Dzamburu]

-q 0.3 is almost 70kbs, and i faac q 100 is about 120kbs correct me if i wrong. 70 is to small for 5.1 encoding and 120 is is normal like you say. So for HE-AAC is only recommend to use 96-192 for LC 192-max. I think in that case and 2pass can't help with that small bitrates

This post has been edited by Dzamburu: May 6 2006, 01:05

[Sagittaire]

-q 0.3 is almost 64kbs correct me if i wrong. This is to small for 5.1 encoding. So for HE-AAC is only recommend to use 96-192 for LC 192-max. I think that and 2pass can't help with that small bitrates

I use LC option encoding. For this source -q 0.3 for neroaacenc done exactly the same bitrate than q 100 for faac. -q 0.3 and -q 100 with this source produce medium bitrate for neroaacenc and faac: 115 Kbps.

[kwanbis]

http://www.hydrogenaudio.org/forums/index....ndpost&p=389146

i mean, why didn't you post on the discussion thread about neroAacEnc?

[Sagittaire]

Well I try with the old nero aac version (aac.dll and aacenc32.dll) and I find that:

profil LC with quality "internet"

for quality "high" size scalling is x2.62
2ch done 2860 Ko
6ch done 7518 Ko

for quality "fast" size scalling is x3.32
2ch done 2713 Ko
6ch done 9022 Ko

for quality "high + pns" size scalling is x2.28
2ch done 2121 Ko
6ch done 4405 Ko

for quality "fast + pns" size scalling is x2.80
2ch done 1727 Ko
6ch done 4842 Ko

Perhaps an old bug for the "fast" preset ... I don't know.

http://www.hydrogenaudio.org/forums/index....ndpost&p=389146

i mean, why didn't you post on the discussion thread about neroAacEnc?

yes I will make that ...

[Garf]

The reason is simply that the -q levels behave a bit differently in 5.1 encoding.

Chanel coupling seem work for 2.0 but not for 5.1.

You are enitrely wrong, channel coupling most definitely works for 5.1.

Perhaps an old bug for the "fast" preset ...

You are entirely wrong, the new encoder has almost no relation to the old ones.

PS. Making these kind of crazy hypotheses each time you don't understand some behaviour of the encoder is silly and annoying.

- for same quality level bitrate for 6ch and 2ch sould be simple scalling something like between x2.0-x2.5

I wonder based on what evidence you're making this statement Do you have access to 5.1 psychoacoustics data that we don't?

What I'm saying is that the intermasking effects of such encodings aren't as well understood yet as we would like, and that you shouldn't be surprised if we manage to drop the bitrate demand of 5.1 encodes very significantly in the future

This post has been edited by Garf: May 6 2006, 09:04

[benc]

The reason is simply that the -q levels behave a bit differently in 5.1 encoding.

I had also noticed that -q levels were giving a higher bit/channel in 5.1 encoding compared to stereo encoding. Does that mean we should use a lower -q for 5.1 encoding than we might otherwise choose to use for mono/stereo encoding?

[Garf]

Yes, that's probably the best thing to do (for now, it will probably change a few more times as we improve 5.1 encoding).

[Dzamburu]

Old Nero 6/7 have most terrible quality in 5.1 encoding for all medium bitrates

I use LC option encoding. For this source -q 0.3 for neroaacenc done exactly the same bitrate than q 100 for faac.

This cant be. Becouse -q 0.3 can use that big bitrate. FAAC -q 100 is eqvivalent nero -q 0.5, 0.6 i think, and for bitrates <192 always use HE mode becouse is better

[Sagittaire]

Making these kind of crazy hypotheses each time you don't understand some behaviour of the encoder is silly and annoying.

Well I make 2 hypotheses ... lol
- Bug for multichanel coupling in 5.1 mode
- quality level are not equivalent

Bug for multichanel or different quality level ... ???

You say that different chanel number imply different quality/chanel with same q level. The other multichanel compatible codec (WMA9 Pro or FAAC for example) seem don't work like that but it's your encoder ... anyway I not make quality accoustic analyse here just bitrate behaviour ... ;-)

I wonder based on what evidence you're making this statement Do you have access to 5.1 psychoacoustics data that we don't?

Me No ... but certainely Yes for Ivan Dimkovic:


['Ivan Dimkovic' post='159185' date='Dec 2 2003, 08:18']

In 5.1 channel config you have following situation:

2 channel pairs (L, R and L', R')
1 single channel ( C )
1 bass, or, LFE channel

In a coding system which does not use so-called inter-channel "coupling" you can figure it out by:

bit rate = 2 x stereo bit rate for same quality + 1 mono bit rate for same q + X, where X is only few kb/s for LFE...

So, for, say AAC - to achieve quality of 128 kb/s, stereo:
2 x 128 + 64 + 1 Kb/s = ~ 320 kb/s very close to, actually 321

For MP3 you would need:
2 x 160 + 80 + x... = ~400 Kb/s for the same quality

For HE-AAC:
2 x 48 + 32 (because of coupling channels need less) = 128 Kb/s

For MP2:
2 x 256 + 128 = 640 Kb/s


Etc...

Anyway if quality level is not the same then analyse is impossible.

This cant be. Becouse -q 0.3 can use that big bitrate. FAAC -q 100 is eqvivalent nero -q 0.5, 0.6 i think, and for bitrates <192 always use HE mode becouse is better

You want my ac3 source ... lol
Be carefull LC, HE and HEV2 don't produce the same range bitrate for same q and I use LC-AAC in this test.


But conclusion is interessing. q level don't produce same real quality by chanel. In this example q 0.3 in 5.1 mode produce very better quality for each chanel than q 0.3 in 2.0 mode.

This post has been edited by Sagittaire: May 6 2006, 13:32

[Dzamburu]

You want my ac3 source ... lol
Be carefull LC, HE and HEV2 don't produce the same range bitrate for same q and I use LC-AAC in this test.


But conclusion is interessing. q level don't produce same real quality by chanel. In this example q 0.3 in 5.1 mode produce very better quality for each chanel than q 0.3 in 2.0.

Hmm that strange, can you give me that ac3 if not to big of course

[Sagittaire]

You want my ac3 source ... lol
Be carefull LC, HE and HEV2 don't produce the same range bitrate for same q and I use LC-AAC in this test.


But conclusion is interessing. q level don't produce same real quality by chanel. In this example q 0.3 in 5.1 mode produce very better quality for each chanel than q 0.3 in 2.0.

Hmm that strange, can you give me that ac3 if not to big of course

useless my encoding is good, if you don't choose LC, HE and HEV2 profil the encoder make itself automatic switch between LC, HE and HEV2. For this source q100 for faac produce same bitrate than q0.3 for neroaacenc in stereo mode.

from faac CLI documentation

-q <quality> Set default variable bitrate (VBR) quantizer quality in percent.

(default: 100, averages at approx. 120 kbps VBR for a normal
stereo input file with 16 bit and 44.1 kHz sample rate; max.
value 500, min. 10).

from neroaacenc CLI documentation

Using VBR mode (-q) will give the best quality. The default setting is -q 0.5, which will give about 160-170kbps on average

This post has been edited by Sagittaire: May 6 2006, 13:11