HydrogenAudio

Lossy Audio Compression => AAC => AAC - General => Topic started by: Inds on 2011-08-14 10:20:50

Title: Why does Nero AAC suck at lower sample rates?
Post by: Inds on 2011-08-14 10:20:50
I use q 0.35 which is damn good quality at 44 or 48 khz but some files that are 44-48 khz don't have frequencies higher than 22 so I downsample to save some space. This is what I used to do with MP3. Since there was audibly no difference in quality between 32 and 44 khz I would encode at 128 kb/s at 32 khz and it would be transparent because it was basically equivalent to 176 kb/s at 44.1 khz.

Anyway, I can't do the same with AAC because q 0.35 with 32 khz has audible artifacts, and q 0.35 with 22.05 khz sounds like GARBAGE. When I increase it to q 0.40 with 32 khz it sounds much better, but the bitrate is only a bit lower than q 0.35 at 44.1 khz which is disappointing. Furthermore, I did a test a long time ago with the same song at different khz with the same bitrate with AAC, and regardless of the same bitrate, ones with lower khz were far lower quality.

ABC/HR Version 1.0, 6 May 2004
Testname: Song at 44/32/22KHz 48kbps AAC

1L = C:\m44(19).wav
2R = C:\m22(35).wav
3L = C:\m32(25).wav

---------------------------------------
General Comments:

---------------------------------------
1L File: C:\m44(19).wav
1L Rating: 3.6
1L Comment: Somewhat distorted, especially on highs.
---------------------------------------
2R File: C:\m22(35).wav
2R Rating: 2.5
2R Comment: Really distorted.
---------------------------------------
3L File: C:\m32(25).wav
3L Rating: 3.3
3L Comment: Slightly worse than sample 1 I think.
---------------------------------------
ABX Results:
Original vs C:\m44(19).wav
    10 out of 10, pval < 0.001
Original vs C:\m22(35).wav
    10 out of 10, pval < 0.001
Original vs C:\m32(25).wav
    10 out of 10, pval < 0.001

The number inside the brackets is the -q setting used, and all songs were 48 kb/s.

Can a developer of Nero confirm this? Is Nero AAC simply not optimized to handle audio with sample rates lower than 44100?
Title: Why does Nero AAC suck at lower sample rates?
Post by: lvqcl on 2011-08-14 10:59:46
I resampled one song from 44.1 KHz to 32 and 22.05 and encoded with Nero -q 0.35, Winamp FhG AAC quality 3 and QT tvbr quality 45.

Bitrates for 44.1 kHz track:
Nero: 100 kbps; FhG: 100 kbps; QT: 90 kbps. More or less the same.

Bitrates for 32 kHz track:
Nero: 65 kbps; FhG: 80 kbps; QT: 90 kbps. Nero produces file with noticeably lower bitrate than two other encoders.

Bitrates for 22.05 kHz track:
Nero: 45 kbps; QT: 60 kbps (encodes as HE-AAC). FhG resamples it to 44.1kHz and encodes at 114 kbps.

Quote
When I increase it to q 0.40 with 32 khz it sounds much better, but the bitrate is only a bit lower than q 0.35 at 44.1 khz which is disappointing.

As you can see, that's how Quicktime (90->90 kbps) and Fraunhofer(100->80 kbps) work.
Title: Why does Nero AAC suck at lower sample rates?
Post by: /mnt on 2011-08-14 11:20:15
I have tried some Super Nintendo rips since that outputs at 32Khz. Also I can confirm that Nero AAC does pretty poor at low sample rates.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1.8 beta 4
2011/08/14 10:47:38

File A: C:\Games\Rips\SPC (SNES)\fzero.rar|Big Blue.spc
File B: C:\Temp\Big Blue.m4a

10:47:38 : Test started.
10:48:31 : 01/01  50.0%
10:48:39 : 02/02  25.0%
10:48:48 : 03/03  12.5%
10:48:55 : 04/04  6.3%
10:49:07 : 05/05  3.1%
10:49:14 : 06/06  1.6%
10:49:27 : 07/07  0.8%
10:49:50 : 08/08  0.4%
10:49:57 : 09/09  0.2%
10:50:08 : 10/10  0.1%
10:50:11 : Test finished.

 ----------
Total: 10/10 (0.1%)
Obvious warbling artifacts.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1.8 beta 4
2011/08/14 10:58:37

File A: C:\Games\Rips\SPC (SNES)\fzero.rar|Port Town.spc
File B: C:\Temp\Port Town.m4a

10:58:37 : Test started.
10:58:51 : 01/01  50.0%
10:58:55 : 02/02  25.0%
10:58:59 : 03/03  12.5%
10:59:04 : 04/04  6.3%
10:59:15 : 05/05  3.1%
10:59:21 : 06/06  1.6%
10:59:27 : 07/07  0.8%
10:59:35 : 08/08  0.4%
10:59:42 : 09/09  0.2%
11:00:05 : 10/10  0.1%
11:00:06 : Test finished.

 ----------
Total: 10/10 (0.1%)
Nero AAC does very poorly with the highs.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1.8 beta 4
2011/08/14 11:03:04

File A: C:\Games\Rips\SPC (SNES)\sf2.rar|Street Fighter 2\Sf2 - 08 - ken.spc
File B: C:\Temp\Sf2 - 08 - ken.m4a

11:03:04 : Test started.
11:03:23 : 01/01  50.0%
11:03:32 : 02/02  25.0%
11:04:14 : 03/03  12.5%
11:04:19 : 04/04  6.3%
11:04:32 : 05/05  3.1%
11:04:43 : 06/06  1.6%
11:04:55 : 07/07  0.8%
11:05:02 : 08/08  0.4%
11:05:09 : 09/09  0.2%
11:05:16 : 10/10  0.1%
11:05:21 : Test finished.

 ----------
Total: 10/10 (0.1%)
Same problem as the other samples.

LAME V7 vs Nero q 0.35
Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1.8 beta 4
2011/08/14 10:53:12

File A: C:\Temp\Big Blue.m4a
File B: C:\Temp\Big Blue.mp3

10:53:12 : Test started.
10:53:22 : 01/01  50.0%
10:53:29 : 02/02  25.0%
10:53:54 : 03/03  12.5%
10:54:02 : 04/04  6.3%
10:54:26 : 05/05  3.1%
10:54:44 : 06/06  1.6%
10:54:51 : 07/07  0.8%
10:54:56 : 08/08  0.4%
10:55:08 : 09/09  0.2%
10:55:24 : 10/10  0.1%
10:55:26 : Test finished.

 ----------
Total: 10/10 (0.1%)
LAME sounds far better then the Nero AAC encode. The bitrate difference is huge though, Nero AAC is 51kbps and LAME V7 was 101kbps.

The bitrates seem to pretty low with Nero AAC, which lvqcl discovered.
Title: Why does Nero AAC suck at lower sample rates?
Post by: lvqcl on 2011-08-14 11:44:23
I tested 2-pass ABR @ 64 kbps (HE-AAC) with "Sabbath Bloody Sabbath" by Cardigans.

Resampling to 32kHz makes encoding better for me.
Title: Why does Nero AAC suck at lower sample rates?
Post by: /mnt on 2011-08-14 16:48:04
I have created an 30 sec sample of a video game track that is decoded from foo_gep at 32KHz. Nero AAC struggles greatly with this sample, I've ABXed it at -q 0.50 without much effort.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1.8 beta 4
2011/08/14 16:20:05

File A: C:\Temp\Mute City.flac
File B: C:\Temp\Mute City.m4a

16:20:05 : Test started.
16:20:36 : 01/01  50.0%
16:20:41 : 02/02  25.0%
16:21:05 : 03/03  12.5%
16:21:14 : 04/04  6.3%
16:21:25 : 05/05  3.1%
16:21:34 : 06/06  1.6%
16:21:42 : 07/07  0.8%
16:21:52 : 08/08  0.4%
16:21:58 : 09/09  0.2%
16:22:08 : 10/10  0.1%
16:22:17 : 11/11  0.0%
16:22:26 : 12/12  0.0%
16:22:27 : Test finished.

 ----------
Total: 12/12 (0.0%)

Sample can be found here (http://www.hydrogenaudio.org/forums/index.php?act=attach&type=post&id=6648).
Title: Why does Nero AAC suck at lower sample rates?
Post by: lvqcl on 2011-08-14 17:38:59
This reminded me about this post: http://www.hydrogenaudio.org/forums/index....st&p=757651 (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=88872&view=findpost&p=757651)
Title: Why does Nero AAC suck at lower sample rates?
Post by: Inds on 2011-08-14 17:57:20
Hey guys, it's great that so many of you are willing to take time to do tests and further validate my observations, but can I get some answers?
Title: Why does Nero AAC suck at lower sample rates?
Post by: DVDdoug on 2011-08-15 20:48:34
Quote
but can I get some answers?
From that information, I guess the best "answer" is it uses a lower bitrate when you feed-in a lower sample rate.  You'd have to ask the developers why it does that, or why it seems to over-compensates.

And the "solution" is probably to use a higher q setting for lower sample-rate files, so that when you can compare "apples to apples" files with approximately the same bitrates (and the same file size).
Title: Why does Nero AAC suck at lower sample rates?
Post by: slks on 2011-10-09 22:24:19
I doubt that Nero AAC has been tuned very well for sample rates other than 44.1 or 48. It is, after all, a Nero product - the developers are probably most focused on audio that's getting encoded to/from CDs or DVDs.