Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Nero AAC Codec 1.3.3.0 (Read 147362 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Nero AAC Codec 1.3.3.0

Reply #75

Can you post the exact command line you used?


I use Foobar which has a graphical interface to the program but selecting custom shows command line options for the previous setting which are (for Q1.00/~400kbps):

-q 1.00 -ignorelength -if - -of %d

The only file that didn't show a 18k lowpass was a 24bit/96khz wav file I also converted at Q 1.00 but the final bit rate was ~850kbps.

Edit: I just did a conversion of Pink Floyd - Money (from WAV) using the following command line in Foobar: 

-br 400000 -ignorelength -if - -of %d

with the same result - a clear 18k lowpass.


I just checked. There is no hardcoded cutoff, what you see is the result of psychoacoustics. The question of course is if this behavior is desirable at such bitrates, but our changes have solved some other high bitrate problems (that were audible). We will investigate this for next version.

BTW: Can you ABX the problem?

Nero AAC Codec 1.3.3.0

Reply #76
I just checked. There is no hardcoded cutoff, what you see is the result of psychoacoustics. The question of course is if this behavior is desirable at such bitrates, but our changes have solved some other high bitrate problems (that were audible). We will investigate this for next version.

BTW: Can you ABX the problem?


With the headphones  I have right now, highly unlikely. It's quite possible, given some of the musical content and my ears, that I wouldn't notice a difference using Senn HD650s, but anyone who chooses to use a bitrate of 400kbps+ should be give the benefit of the doubt. 

Nero AAC Codec 1.3.3.0

Reply #77
There is no hardcoded cutoff, what you see is the result of psychoacoustics. The question of course is if this behavior is desirable at such bitrates, but our changes have solved some other high bitrate problems (that were audible). We will investigate this for next version.
With the headphones  I have right now, highly unlikely. It's quite possible, given some of the musical content and my ears, that I wouldn't notice a difference using Senn HD650s, but anyone who chooses to use a bitrate of 400kbps+ should be give the benefit of the doubt. 

As Menno wrote, psychoacoustics choses here to use available bits on other parts of spectrum. There is no cutoff. Using those bits on content above 18kHz that is not expected to be heard, doesn't mean that quality will be better. It is only about what you hear and not what spectrum looks like.

Nero AAC Codec 1.3.3.0

Reply #78
For now I'll keep using the Windows binary with Wine instead of the native Linux binary


Yeah, similar results here.  The win32 binary takes about 3/4 as long as the native linux binary on a 6 channel S24LE input wav file. (c2d e6600).  They're both 32bit binaries , and the Linux one looks like it might well have been compiled with gcc.  Yeah, strings -a on the linux binary shows a few instances of
GCC: (GNU) 4.2.1 (Ubuntu 4.2.1-5ubuntu4)
and many of
GCC: (GNU) 4.1.3 20070929 (prerelease) (Ubuntu 4.1.2-16ubuntu2)
and some code assembled with
Microsoft ® Macro Assembler
Dc:\temp\ipptmp_20688\ipptmp_5224\psdp32fv8as\psdp32fv8as_nosplit.obj4
...

The important difference is that the win32 binary makes significant use of SSE2 instructions that operate on packed double-precision floats (e.g. mulpd), while the Linux binary doesn't.  It does have some sse2 code kicking around (e.g. cvtps2pd), but no usage of mulpd.  It uses mulsd in a few places, probably gcc-generated.  If the vector double code was generated by the compiler they used on Windows, then yeah, maybe using the Intel compiler for Linux could help.  If it's hand-written asm, then they're just not finished porting their asm optimizations to Linux yet.

BTW, I just used objdump -d linux/neroAacEnc |less  and  i586-mingw32msvc-objdump -d win32/neroAacEnc.exe | less
and searched for "pd %" to find instructions with names ending in pd.  (pd = packed double [precision float].  sd = scalar double, ps = packed single.)

Nero AAC Codec 1.3.3.0

Reply #79
Hi,guys
I'v found the same question as "Sartre" said there is a cutoff at the frequency most probably between 16KHz to 18Khz with neroAacEnc.exe 1.3.3.0 whatever the -q setting even at q=1.

And I'v found more that this new release 1.3.3.0 can't set -lc mode.When you set command e.g.
-q 1 -lc -ignorelength -if - -of %d
The result that adobe audition(1.5 with faac&id3lib) shows always AAC HE.

First time,I realize maybe id3lib was too old,but checked it was newest.I also read about faac that only LC mode maybe it can't decode HE AAC that lose the high frequency.

Then I install nero 6.0 and nero aac for audition(Nero MPEG-4 filter 0.0.3).That include aac.dll(2.5.9.991) aacenc32.dll(3.2.0.20).
And use this plugin to translate the same .flac sample to nero aac HE and/or LC mode that show no longer cutoff frequency and keep it perfect spectrum from highest solution(250~300kbps) to even low HE_aac at 30~40kbps.
Indeed,using faac&id3lib filter or nero aac plugin filter separately can both recognize HE and LC mode made by nero aac plugin for audition but not neroAacEnc.exe 1.3.3.0 made.

So,whatever you set command -lc or leave it as default -he ,neroAacEnc.exe 1.3.3.0  always produces only HE_AAC,and also its lowpass cutoff problem,is that so?

ps: I  certainly agree "muaddib" said "It is only about what you hear and not what spectrum looks like",but as the equipments go better that lowpass cutoff too low will affect the quality.

Nero AAC Codec 1.3.3.0

Reply #80
HE AAC is definately not used at -q 1.0, your decoding software simply shows it incorrectly. Old versions of FAAD2 reports HE explicitly turned off incorrectly (as turned on).

 

Nero AAC Codec 1.3.3.0

Reply #81
HE AAC is definately not used at -q 1.0, your decoding software simply shows it incorrectly. Old versions of FAAD2 reports HE explicitly turned off incorrectly (as turned on).


I'v tried -q 0.3 to 1.0 too.

Is there a way to detect whether HE_AAC or LC_AAC ?

Nero AAC Codec 1.3.3.0

Reply #82
FAAD2 has been bugfixed, but you will need to find a recent compile. Foobar2000 also shows the information correctly AFAIK

Nero AAC Codec 1.3.3.0

Reply #83
FAAD2 has been bugfixed, but you will need to find a recent compile. Foobar2000 also shows the information correctly AFAIK


foobar2000 shows out its codec profile just right as encoder set.
FAAD2 ,see whether I can compile one and then try it again.

But,the lowpass cutoff now should be confirmed for now I have tested with nero 9.2.5.0 nero_WaveEditor.
It shows the same result as AA gives.
Maybe I guess that the new release keeps the bitrates a priority at energy focus place or quantification?
that's weird every convert results the same cutoff frequency 18KHz and some 16KHz if there is low energy between 16KHz and 18KHz.

Nero AAC Codec 1.3.3.0

Reply #84
ps: I  certainly agree "muaddib" said "It is only about what you hear and not what spectrum looks like",but as the equipments go better that lowpass cutoff too low will affect the quality.

"The equipments" is already good enough. It is your (and mine) ears that can not detect the whole spectrum and for sure this will not improve as we grow older. And it is not only because somebody can not hear some high frequencies but also because higher frequencies are masked by signal from lower frequencies. As I already wrote, it is important what you hear and not what music looks like in a spectrum analyzer. Please do ABX test and if you prove that you hear difference then please report the sample for which it is the case.

Nero AAC Codec 1.3.3.0

Reply #85
We hope that you'll give new encoder a try even if you don't get iPod control unit
Your findings were very helpful.


I know this is an old post but I didn't see it when it was posted and I always try to reply when a developer takes the time to directly reply to one of my posts.

I purchased the iPod integration kit today and it is great.  My main concern was using my Civic's auxiliary port which produces inferior results when compared to an mp3 CD, WMA CD, and standard audio CD.  The bass would increase and the highs would be muddled.  I tried changing the EQ on the deck and on my iPod with no luck.  I guess an analog connection, in my car, just can't compare to a digital connection.

I was able to find Honda's iPod connectivity kit and get it installed by the dealership for $150 in all, not too bad.  I have an old 4G 40GB iPod that I now keep permanently in my car.  The iPod kit uses a digital connection for the audio.  It uses the same connection as the 6 disc changer that goes in the trunk.  I cannot hear any noticeable differences between playing a mp3 file on a data disc and through the iPod connection kit (it is really hard to do it blindly but I am still satisfied).

So I am now switching to the new Nero AAC encoder at the -q0.40 setting.  foobar2000 will be encoding my lossless library overnight (for three total days of encoding).  I just wish that my PS3 supported the iTunes/Apple tagging standard.  Oh well, at least the files (and track tags) will be displayed on my Xbox 360 and Apple TV.  I prefer to use my Apple TV for media streaming but there are times when I want to use my PS3 due to my PS3 being hooked up to a different system.

I conducted blind ABX tests and think that the -q0.40 setting is too much for my ears but the files produced come out to around the same file size at 128kbps VBR iTunes AAC files (sometimes the files are 0.1 MB higher and sometimes they are 0.1MB lower).  All-in-all I am really pleased with the new encoder.

Nero AAC Codec 1.3.3.0

Reply #86
I just wish that my PS3 supported the iTunes/Apple tagging standard


...and this is really ridiculous from Sony and his "flagship" since their Walkman portable players and even the PSP read the AAC tags fine. I hope to not wait PS4...
Anyway, even the X360 is in the same situation so i can't understand for both why support a format at half... 

Nero AAC Codec 1.3.3.0

Reply #87
Xbox 360 can read AAC tags when an iPod or Zune are plugged into it and when I use media streaming software such as the Zune software.  They are on the right path.  AAC tags still have issues when reading from other USB devices and data CDs/DVDs.  This is still better than the PS3 as I either have to go through the painful process of converting my tags and putting my AAC files in a 3GP container or I can use a media streaming application that will transcode the files on-the-fly.

Nero AAC Codec 1.3.3.0

Reply #88
This is still better than the PS3...


I don't think because the worse thing of X360 is that doesn't has only tags issues but also doesn't play all the files... I tried 2 iTunes+ (so no DRM) albums that totally refuse to play. PS3 play them fine...

Nero AAC Codec 1.3.3.0

Reply #89
Tagging standards are a pain.   

kornchild2002: Are you having any gapless playback issues on these devices?  I keep running into snags with gapless playback on Nero files (the last three releases) unless I transcode using Foobar.  If I use Dbpoweramp's converter or even ripping straight from EAC, I find gapless issues in Winamp and Rockbox.  iTunes-encoded AAC is still gapless.  It is frustrating, because Dark Side of the Moon, Wish You Were Here, The Fragile, etc. are painful when you get kicked in that stomach unexpectedly for that nanosecond of a song transition!  I think gapless playback moves up to my top priority in choosing a codec.  Nero, from what I understand, is fully gapless-compliant, but that only is as good as the encoding process used and then the decoder/player support.

Nero AAC Codec 1.3.3.0

Reply #90
I don't have any gapless issues with Nero AAC.  Then again, I use iTunes to sync my AAC files to my iPod.  iTunes has to scan the files for gapless playback first but I don't have any issues after that.

Nero AAC Codec 1.3.3.0

Reply #91
I found another killer sample that Nero AAC really struggles on. Which IMO sounds like a Blade encoding at q 0.40 - 0.55:

Nero AAC 1.3.3.0 q 0.50

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/16 21:17:58

File A: C:\Temp\Linchpin [Edit].flac
File B: C:\Temp\Linchpin.mp4

21:17:58 : Test started.
21:18:22 : 01/01  50.0%
21:18:34 : 02/02  25.0%
21:18:47 : 03/03  12.5%
21:19:04 : 04/04  6.3%
21:19:18 : 05/05  3.1%
21:19:31 : 06/06  1.6%
21:19:43 : 07/07  0.8%
21:19:49 : 08/08  0.4%
21:19:57 : 09/09  0.2%
21:20:07 : 10/10  0.1%
21:20:23 : 11/11  0.0%
21:20:38 : 12/12  0.0%
21:20:39 : Test finished.

 ----------
Total: 12/12 (0.0%)

Warbling and smear. Very easy to ABX

q 0.55

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/16 21:21:37

File A: C:\Temp\Linchpin [Edit].flac
File B: C:\Temp\Linchpin q 0.55.m4a

21:21:37 : Test started.
21:21:53 : 01/01  50.0%
21:22:06 : 02/02  25.0%
21:22:19 : 02/03  50.0%
21:22:25 : 03/04  31.3%
21:22:35 : 04/05  18.8%
21:22:39 : 05/06  10.9%
21:22:51 : 06/07  6.3%
21:23:09 : 06/08  14.5%
21:23:14 : 07/09  9.0%
21:23:24 : 07/10  17.2%
21:23:28 : 08/11  11.3%
21:23:32 : 09/12  7.3%
21:23:40 : 10/13  4.6%
21:23:45 : 11/14  2.9%
21:23:54 : 12/15  1.8%
21:24:06 : 13/16  1.1%
21:24:14 : 14/17  0.6%
21:24:18 : 15/18  0.4%
21:24:23 : 16/19  0.2%
21:24:30 : 17/20  0.1%
21:24:45 : 18/21  0.1%
21:24:53 : 19/22  0.0%
21:24:55 : Test finished.

 ----------
Total: 19/22 (0.0%)

Smearing gone; but some warbling on the guitar. Hard to ABX.

The sample is transparent with Quicktime VBR AAC; I posted the logs on this topic, also rpp3po can ABX this sample aswell.
"I never thought I'd see this much candy in one mission!"

Nero AAC Codec 1.3.3.0

Reply #92
Thanks for reporting it. Taken into account for the next release.

Nero AAC Codec 1.3.3.0

Reply #93
I too confirm version 1.3.3.0 problem appears to apply a 18k (or lower) lowpass on everything, regardless of bitrate.
In version 1.1.34.2 of such problems was not.

Whether it is planned to use technology NVidia CUDA for speed increase in tens times? 

Nero AAC Codec 1.3.3.0

Reply #94
I too confirm version 1.3.3.0 problem appears to apply a 18k (or lower) lowpass on everything, regardless of bitrate.
In version 1.1.34.2 of such problems was not.

Whether it is planned to use technology NVidia CUDA for speed increase in tens times? 


"Low-pass" that you noticed was already discussed and is not fixed low-pass but product of psychoacoustic analysis on specific samples during encoding.
Do you hear on any file this reduction of frequencies above 18kHz? Can you prove it with ABX? If you find any sample that you can ABX because of >18kHz reduction, can you try that sample with 1.1.34.2 and try to ABX it with that version? If you don't hear differences with 1.1.34.2 for that sample. can you then apply 18kHz low-pass to decoded output and try to ABX it then?

NVidia CUDA wouldn't improve speed much. Too complicated to implement for small speed improvements for small number of people with powerful enough PCs.

Nero AAC Codec 1.3.3.0

Reply #95
I have a few more samples that produce noticable artifacts at low bitrates.

I know that -q 30 is far from transparent, but that setting seems to really struggle with vocals on this sample:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/05 21:41:14

File A: C:\Temp\Birthday Dethday [Edit] -q 30.mp4
File B: C:\Temp\Birthday Dethday [Edit].flac

21:41:14 : Test started.
21:41:27 : 01/01  50.0%
21:41:34 : 02/02  25.0%
21:41:44 : 03/03  12.5%
21:41:50 : 04/04  6.3%
21:41:56 : 05/05  3.1%
21:42:01 : 06/06  1.6%
21:42:07 : 07/07  0.8%
21:42:13 : 08/08  0.4%
21:42:19 : 09/09  0.2%
21:42:24 : 10/10  0.1%
21:42:30 : 11/11  0.0%
21:42:35 : 12/12  0.0%
21:42:39 : Test finished.

 ----------
Total: 12/12 (0.0%)

Underwater volcals.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/05 21:43:20

File A: C:\Temp\Birthday Dethday [Edit] -q 30.mp4
File B: C:\Temp\Birthday Dethday [Edit] V8.mp3

21:43:20 : Test started.
21:43:32 : 01/01  50.0%
21:43:38 : 02/02  25.0%
21:43:48 : 03/03  12.5%
21:43:54 : 04/04  6.3%
21:43:59 : 05/05  3.1%
21:44:07 : 06/06  1.6%
21:44:15 : 07/07  0.8%
21:44:19 : 08/08  0.4%
21:44:25 : 09/09  0.2%
21:44:31 : 10/10  0.1%
21:44:32 : Test finished.

 ----------
Total: 10/10 (0.1%)

Vocals are lot more clearer on the LAME V8 encode. IMO the LAME V8 encode sounds better, even if the drums sound worse.

Also I found a noticable artifact on start of a track at -q 0.50:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/05 21:13:40

File A: F:\Music\Albums\Skinny Puppy - Too Dark Park\01. Convulsion.flac
File B: C:\Temp\Convulsion (Nero AAC 1.3.3.0 -q 0.50).m4a

21:13:40 : Test started.
21:14:06 : 01/01  50.0%
21:14:11 : 02/02  25.0%
21:14:18 : 03/03  12.5%
21:14:29 : 04/04  6.3%
21:14:35 : 05/05  3.1%
21:14:44 : 06/06  1.6%
21:14:52 : 07/07  0.8%
21:15:04 : 08/08  0.4%
21:15:16 : 09/09  0.2%
21:15:26 : 10/10  0.1%
21:15:33 : 11/11  0.0%
21:15:40 : 11/12  0.3%
21:15:43 : 12/13  0.2%
21:15:49 : 13/14  0.1%
21:15:54 : 14/15  0.0%
21:16:06 : 15/16  0.0%
21:16:14 : 16/17  0.0%
21:16:20 : Test finished.

 ----------
Total: 16/17 (0.0%)

Artifact appears around at 0:04.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/05 21:17:20

File A: F:\Music\Albums\Skinny Puppy - Too Dark Park\01. Convulsion.flac
File B: C:\Temp\Convulsion (Nero AAC 1.3.3.0 -q 0.55).m4a

21:17:20 : Test started.
21:17:51 : 00/01  100.0%
21:17:55 : 01/02  75.0%
21:18:03 : 02/03  50.0%
21:18:10 : 03/04  31.3%
21:18:22 : 04/05  18.8%
21:18:36 : 04/06  34.4%
21:18:45 : 04/07  50.0%
21:18:48 : 05/08  36.3%
21:18:53 : 05/09  50.0%
21:19:00 : 06/10  37.7%
21:19:06 : 07/11  27.4%
21:19:12 : 08/12  19.4%
21:19:31 : 08/13  29.1%
21:19:38 : 08/14  39.5%
21:19:42 : 08/15  50.0%
21:19:47 : 09/16  40.2%
21:19:50 : 10/17  31.5%
21:19:58 : 11/18  24.0%
21:20:03 : 12/19  18.0%
21:20:12 : 13/20  13.2%
21:20:21 : 14/21  9.5%
21:20:27 : 15/22  6.7%
21:20:33 : 16/23  4.7%
21:20:39 : 17/24  3.2%
21:20:43 : 17/25  5.4%
21:20:49 : 18/26  3.8%
21:20:55 : 19/27  2.6%
21:21:01 : 19/28  4.4%
21:21:03 : Test finished.

 ----------
Total: 19/28 (4.4%)

I think its still almost there, but i gave up.

Samples:

Birthday Dethday
Convulsion
"I never thought I'd see this much candy in one mission!"

Nero AAC Codec 1.3.3.0

Reply #96
The Convulsion sample is also ABXable with LAME 3.98.2 at V0:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/06 01:12:37

File A: C:\Temp\Convulsion [Edit] V0.mp3
File B: C:\Temp\Convulsion [Edit].flac

01:12:37 : Test started.
01:12:55 : 01/01  50.0%
01:13:05 : 02/02  25.0%
01:13:11 : 02/03  50.0%
01:13:15 : 03/04  31.3%
01:13:20 : 04/05  18.8%
01:13:26 : 05/06  10.9%
01:13:34 : 06/07  6.3%
01:13:41 : 07/08  3.5%
01:13:49 : 08/09  2.0%
01:13:56 : 09/10  1.1%
01:14:19 : 10/11  0.6%
01:14:24 : 11/12  0.3%
01:14:30 : 11/13  1.1%
01:14:34 : 12/14  0.6%
01:14:40 : 13/15  0.4%
01:14:48 : 14/16  0.2%
01:15:01 : 15/17  0.1%
01:15:09 : 16/18  0.1%
01:15:16 : 17/19  0.0%
01:15:21 : 18/20  0.0%
01:15:29 : 19/21  0.0%
01:15:38 : 19/22  0.0%
01:15:52 : 20/23  0.0%
01:16:02 : 21/24  0.0%
01:16:08 : 22/25  0.0%
01:16:12 : Test finished.

 ----------
Total: 22/25 (0.0%)

Precho hits at over 0:04.
"I never thought I'd see this much candy in one mission!"

Nero AAC Codec 1.3.3.0

Reply #97
Thanks for the report, we will look into it.
At -q 0.3 the encoder uses SBR, I guess this is a sample that sounds worse with SBR than without (around that bitrate at least).

Nero AAC Codec 1.3.3.0

Reply #98
Thanks for the report, we will look into it.
At -q 0.3 the encoder uses SBR, I guess this is a sample that sounds worse with SBR than without (around that bitrate at least).

I've tested it with -q 0.3 -lc switch, which uses the LC profile, and it does seem to be a SBR issue.

-q 0.30 -lc

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/06 01:56:01

File A: C:\Temp\Birthday Dethday [Edit].flac
File B: C:\Temp\Birthday Dethday [Edit].m4a

01:56:01 : Test started.
01:56:12 : 01/01  50.0%
01:56:21 : 02/02  25.0%
01:56:26 : 03/03  12.5%
01:56:33 : 04/04  6.3%
01:56:45 : 05/05  3.1%
01:56:55 : 06/06  1.6%
01:57:02 : 07/07  0.8%
01:57:10 : 08/08  0.4%
01:57:17 : 09/09  0.2%
01:57:22 : 10/10  0.1%
01:57:24 : Test finished.

 ----------
Total: 10/10 (0.1%)

Drums sound bad, just like the LAME V8 encode; but the vocals are clear.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/06 01:57:55

File A: C:\Temp\Birthday Dethday [Edit] -q 30.mp4
File B: C:\Temp\Birthday Dethday [Edit].m4a

01:57:55 : Test started.
01:58:10 : 01/01  50.0%
01:58:13 : 02/02  25.0%
01:58:16 : 03/03  12.5%
01:58:24 : 04/04  6.3%
01:58:28 : 05/05  3.1%
01:58:30 : 06/06  1.6%
01:58:37 : 07/07  0.8%
01:58:40 : 08/08  0.4%
01:58:44 : 09/09  0.2%
01:58:48 : 10/10  0.1%
01:58:59 : 11/11  0.0%
01:59:02 : 12/12  0.0%
01:59:04 : Test finished.

 ----------
Total: 12/12 (0.0%)

Drums on the SBR version sound better, but the vocals sound awful. I prefer the LC version.
"I never thought I'd see this much candy in one mission!"

Nero AAC Codec 1.3.3.0

Reply #99
Maybe it would be usefull to see this sample too.
http://www.hydrogenaudio.org/forums/index....ost&id=5081

q 0.41 is ABXable. There is some background warm noise and as effect snare drum sounds sandy. 

Code: [Select]
ABC/HR Version 1.1 beta 2, 18 June 2004
Testname:

1R = C:\samples\Grip Inc Drums\nero_130 .wav

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

---------------------------------------
ABX Results:
Original vs C:\samples\Grip Inc Drums\nero_130 .wav
    5 out of 5, pval = 0.031