The Pre-echo Machine, ABXing Kraftwerk's The Man Machine at 320kbps CBR Mp3 Options

[/mnt]

I did a very small listening test with LAME at V0 and 320kbps CBR recently and found alot of problems with Kraftwerk's The Man Machine. I know The Robots has been used as a killer sample for many years, but I cannot belive that I managed to successfully ABXed 4/6 tracks from this album.

01. The Robots (LAME 3.98.2 320kbps CBR)

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 4
2009/12/20 00:47:58

File A: C:\Rips\Kraftwerk - The Man Machine (2009 Remaster)\01. The Robots.flac
File B: C:\Temp\The Man Machine ABX Test\The Robots 320 CBR.mp3

00:47:58 : Test started.
00:48:14 : 01/01 50.0%
00:48:19 : 02/02 25.0%
00:48:27 : 03/03 12.5%
00:48:32 : 04/04 6.3%
00:48:40 : 05/05 3.1%
00:48:47 : 06/06 1.6%
00:48:54 : 07/07 0.8%
00:48:59 : 08/08 0.4%
00:49:04 : 09/09 0.2%
00:49:10 : 10/10 0.1%
00:49:16 : 11/11 0.0%
00:49:22 : 12/12 0.0%
00:49:28 : 13/13 0.0%
00:49:33 : 14/14 0.0%
00:49:38 : 15/15 0.0%
00:49:44 : 16/16 0.0%
00:49:54 : 17/17 0.0%
00:49:59 : 18/18 0.0%
00:50:06 : 19/19 0.0%
00:50:11 : 20/20 0.0%
00:50:12 : Test finished.

----------
Total: 20/20 (0.0%)

This is a well known killer track that needs no introduction.

02. Spacelab (LAME 3.98.2 320kbps CBR)

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 4
2009/12/20 00:51:03

File A: C:\Rips\Kraftwerk - The Man Machine (2009 Remaster)\02. Spacelab.flac
File B: C:\Temp\The Man Machine ABX Test\Spacelab 320 CBR.mp3

00:51:03 : Test started.
00:51:45 : 01/01 50.0%
00:51:55 : 02/02 25.0%
00:52:04 : 03/03 12.5%
00:52:12 : 04/04 6.3%
00:52:21 : 05/05 3.1%
00:52:30 : 06/06 1.6%
00:52:39 : 07/07 0.8%
00:52:49 : 08/08 0.4%
00:52:57 : 09/09 0.2%
00:53:05 : 10/10 0.1%
00:53:14 : 11/11 0.0%
00:53:24 : 12/12 0.0%
00:53:47 : 13/13 0.0%
00:53:57 : 14/14 0.0%
00:54:02 : 15/15 0.0%
00:54:04 : Test finished.

----------
Total: 15/15 (0.0%)

Shares the same problem with The Robots (lots of "pfft" pre-echo artifacts), at 1:48 - 1:52 pre-echo is very noticable before the vocoder vocals appear again.

03. Metropolis (LAME 3.98.2 320kbps CBR)

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 4
2009/12/20 00:54:57

File A: C:\Rips\Kraftwerk - The Man Machine (2009 Remaster)\03. Metropolis.flac
File B: C:\Temp\The Man Machine ABX Test\Metropolis 320 CBR.mp3

00:54:57 : Test started.
00:55:18 : 01/01 50.0%
00:55:21 : 02/02 25.0%
00:55:24 : 03/03 12.5%
00:55:27 : 04/04 6.3%
00:55:30 : 05/05 3.1%
00:55:33 : 06/06 1.6%
00:55:36 : 07/07 0.8%
00:55:39 : 08/08 0.4%
00:55:42 : 09/09 0.2%
00:55:46 : 10/10 0.1%
00:55:51 : 11/11 0.0%
00:55:54 : 12/12 0.0%
00:55:57 : 13/13 0.0%
00:56:01 : 14/14 0.0%
00:56:04 : 15/15 0.0%
00:56:07 : 16/16 0.0%
00:56:10 : 17/17 0.0%
00:56:17 : 18/18 0.0%
00:56:20 : 19/19 0.0%
00:56:23 : 20/20 0.0%
00:56:25 : 21/21 0.0%
00:56:28 : 22/22 0.0%
00:56:31 : 23/23 0.0%
00:56:35 : 24/24 0.0%
00:56:40 : 25/25 0.0%
00:56:43 : 26/26 0.0%
00:56:46 : 27/27 0.0%
00:56:51 : 28/28 0.0%
00:56:55 : 29/29 0.0%
00:56:59 : 30/30 0.0%
00:57:00 : Test finished.

----------
Total: 30/30 (0.0%)

Has a really horrid pre-echo / warbling artifact at 0:09 - 0:10.

06. The Man Machine (LAME 3.98.2 320kbps CBR)

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 4
2009/12/20 00:58:06

File A: C:\Rips\Kraftwerk - The Man Machine (2009 Remaster)\06. The Man Machine.flac
File B: C:\Temp\The Man Machine ABX Test\The Man Machine 320 CBR.mp3

00:58:06 : Test started.
00:58:28 : 01/01 50.0%
00:58:38 : 02/02 25.0%
00:58:42 : 03/03 12.5%
00:58:47 : 04/04 6.3%
00:58:52 : 05/05 3.1%
00:59:08 : 06/06 1.6%
00:59:13 : 07/07 0.8%
00:59:18 : 08/08 0.4%
00:59:21 : 09/09 0.2%
00:59:25 : 10/10 0.1%
00:59:39 : 11/11 0.0%
00:59:44 : 12/12 0.0%
00:59:47 : 13/13 0.0%
00:59:53 : 14/14 0.0%
00:59:57 : 15/15 0.0%
00:59:58 : Test finished.

----------
Total: 15/15 (0.0%)

Pre-echo artifact at 0:04.

Almost the whole album itself is riddled with artifacts that Mp3 chokes on greatly, mainly due to its limitations.

At the moment am having a much harder time deciding if i prefer The Man Machine over Trans Europe Express.

This post has been edited by /mnt: Dec 21 2009, 00:13

[shadowking]

So how is AAC nero Q0.5 ? I was happy with 170k MPC quality at the time and also with wv lossy > 300k

[MostlyHarmless]

So how is AAC nero Q0.5 ? I was happy with 170k MPC quality at the time and also with wv lossy > 300k

If I were you, I would stay with MPC, moreover it produce smaller files than Lame (musepack --standard vs. Lame -V2).
I'm a bit surprized that musepack is so good at these bitrates with that type of music, though. IIRC, it was optimized for more 'natural' sounds such as classic and rock music.
For such music I've had to switch from --standard to --extreme just a few times.

This post has been edited by MostlyHarmless: Dec 21 2009, 10:49

[shadowking]

So how is AAC nero Q0.5 ? I was happy with 170k MPC quality at the time and also with wv lossy > 300k

If I were you, I would stay with MPC, moreover it produce smaller files than Lame (musepack --standard vs. Lame -V2).
I'm a bit surprized that musepack is so good at these bitrates with that type of music, though. IIRC, it was optimized for more 'natural' sounds such as classic and rock music.
For such music I've had to switch from --standard to --extreme just a few times.

MPC has nearly no preecho being subband based and suffers different artifact (usually subtle).

ATM - I use lossless for archive and mp3 for listening (v3) and i can't get full transparent sound but quality is good . The thing that annoys me with mp3 is that artifacts can be ugly even if subtle and boosting bitrate offers marginal results. Then SFB21 , -Y etc still causes confusion even today. I don't care too much for the most extreme sample but don't want to abx stuff randomly. You change codec and lose portability. I could use 2 libraries - HQ lossy (even lossless) on PC and a V5 / 4 mp3 archive for devices.

This post has been edited by shadowking: Dec 21 2009, 11:27

[/mnt]

Samples can be found here.

So how is AAC nero Q0.5 ? I was happy with 170k MPC quality at the time and also with wv lossy > 300k

Nero AAC at 0.50 performs alot better then a 320kbps encoded Mp3 at half the size. Some tracks are still ABXable though, but i have to put more effort towards ABXing with AAC. Sadly iTunes fails badly with The Robots, even with the iTunes Plus default setting.

I am also very impressed with the quality that a 170kbps MPC file can produce with heavy pre-echo tracks.

02. Spacelab (Nero AAC 1.3.3.0 -q 0.5)

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 4
2009/12/21 12:21:15

File A: C:\Rips\Kraftwerk - The Man Machine (2009 Remaster)\02. Spacelab.flac
File B: C:\Music\Kraftwerk\The Man Machine (2009 Remaster)\02. Spacelab.m4a

12:21:15 : Test started.
12:21:41 : 01/01 50.0%
12:21:52 : 02/02 25.0%
12:21:57 : 03/03 12.5%
12:22:03 : 04/04 6.3%
12:22:09 : 04/05 18.8%
12:22:19 : 05/06 10.9%
12:22:37 : 06/07 6.3%
12:22:51 : 07/08 3.5%
12:22:59 : 08/09 2.0%
12:23:30 : 09/10 1.1%
12:23:45 : 10/11 0.6%
12:24:17 : 11/12 0.3%
12:24:37 : 12/13 0.2%
12:24:44 : 13/14 0.1%
12:24:57 : 14/15 0.0%
12:24:58 : Test finished.

----------
Total: 14/15 (0.0%)

Smearing at 1:37, not as annoying as most Mp3 encodings i've heard though.

This post has been edited by /mnt: Dec 21 2009, 14:30

[Canar]

Thanks /mnt. Your listening tests are always intriguing, though I often can't ABX what is obvious to you.

[/mnt]

I tried out QuickTimes's true VBR mode this week, with The Robots at --tvbr 100:

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 6
2010/01/08 23:19:22

File A: C:\Rips\Kraftwerk - The Man Machine (2009 Remaster)\01. The Robots.flac
File B: C:\Temp\The Robots (QuickTime VBR 100).m4a

23:19:22 : Test started.
23:20:40 : 01/01 50.0%
23:20:52 : 02/02 25.0%
23:21:09 : 03/03 12.5%
23:21:17 : 04/04 6.3%
23:21:37 : 05/05 3.1%
23:21:46 : 06/06 1.6%
23:22:01 : 07/07 0.8%
23:22:07 : 08/08 0.4%
23:22:14 : 09/09 0.2%
23:22:20 : 10/10 0.1%
23:22:28 : 11/11 0.0%
23:22:34 : 12/12 0.0%
23:22:41 : 13/13 0.0%
23:22:48 : 14/14 0.0%
23:22:56 : 15/15 0.0%
23:22:59 : Test finished.

----------
Total: 15/15 (0.0%)

A obvious artifact appears at 0:30 - 0:33. iTunes and QuickTime performs quitely badly with this track at bitrates below 224kbps. At least the track is transparent to me at --tvbr 127.

Also i found a yet another Kraftwerk track to be ABXable at 320kbps with LAME:

320 CBR

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 6
2010/01/09 16:27:40

File A: C:\Rips\Kraftwerk - Trans Europe Express (2009 Remaster)\03. Showroom Dummies.flac
File B: C:\Temp\Showroom Dummies 320 CBR.mp3

16:27:40 : Test started.
16:28:08 : 01/01 50.0%
16:28:22 : 02/02 25.0%
16:28:32 : 03/03 12.5%
16:28:46 : 04/04 6.3%
16:28:51 : 05/05 3.1%
16:29:03 : 06/06 1.6%
16:29:27 : 07/07 0.8%
16:29:34 : 08/08 0.4%
16:29:44 : 09/09 0.2%
16:30:07 : 10/10 0.1%
16:30:18 : 11/11 0.0%
16:30:30 : 12/12 0.0%
16:30:40 : Test finished.

----------
Total: 12/12 (0.0%)

Smearing at 2:02 - 2:04.

V0

CODE

foo_abx 1.3.4 report
foobar2000 v1.0 beta 6
2010/01/08 23:04:16

File A: C:\Rips\Kraftwerk - Trans Europe Express (2009 Remaster)\03. Showroom Dummies.flac
File B: C:\Temp\Showroom Dummies LAME 3.98.2 V0.mp3

23:04:16 : Test started.
23:05:09 : 00/01 100.0%
23:05:24 : 01/02 75.0%
23:05:37 : 02/03 50.0%
23:05:44 : 03/04 31.3%
23:05:58 : 04/05 18.8%
23:06:15 : 05/06 10.9%
23:06:32 : 06/07 6.3%
23:06:50 : 07/08 3.5%
23:07:07 : 08/09 2.0%
23:07:14 : 09/10 1.1%
23:07:51 : 10/11 0.6%
23:08:05 : 11/12 0.3%
23:08:54 : 12/13 0.2%
23:09:20 : 13/14 0.1%
23:09:26 : 14/15 0.0%
23:09:34 : 14/16 0.2%
23:09:51 : 15/17 0.1%
23:09:58 : 16/18 0.1%
23:10:14 : 17/19 0.0%
23:10:30 : 18/20 0.0%
23:10:38 : 19/21 0.0%
23:10:45 : 20/22 0.0%
23:10:48 : Test finished.

----------
Total: 20/22 (0.0%)

Smearing at 2:02 - 2:04.

Also the smearing artifact appears on V2 and it barely improves at 320kbps.

[DeepSEA]

I'm curious how these tracks do with Ogg Vorbis. This is one of my all time favorite albums! Maybe its time to stop lurking and start ABXing...

BTW, this is my first post! I have gotten tons of good info from this site over the years!

Scott

This post has been edited by DeepSEA: Feb 2 2011, 15:46

[halb27]

I use a personal version of Lame 3.98.4 in order to maximize quality of Lame VBR. This applies to the accuracy especially of short blocks which are used when Lame encounters pre-echo prone music (though I personally do it for other reasons more than because of pre-echo issues which I'm pretty insensitive to).
I use it with the options -V0 -b 320 -F --ns-bass -9 --ns-alto -7 --ns-treble -6 -Y --noreplaygain (and apply the lossless mp3packer procedure afterwards to regain some bitrate).
A negative value of --ns-bass/alto/treble makes Lame more defensive, and the only deviation of my version from the orginal lies in the details of these parameters.

If you like to you can try this especially defensive way of using Lame VBR by using the link above.
With especially pre-echo problems in mind it might be advantageous to reduce the absolute value of --ns-bass, for instance to --ns-bass -6, and increase the absolute value of --ns-alto and --ns-treble, for instance to --ns-alto -8 --ns-treble -8.

This post has been edited by halb27: Feb 2 2011, 23:14

[Canar]

The Internet where I am is being irritatingly flaky, so I'm strugging to search the forums for them. In the future please back up such claims about audio quality with references to your blind tests, halb27, as per our Terms of Service.

[halb27]

I ABXed improvements with (small) tremolo issues of plain -V0 on 'lead-voice', and with (small) harpsichord issues of plain -V0 on 'harp40_1'.
Moreover with this usage I was not able to ABX a (tiny) issue with french female voice on 'Là oú je suis née' I once could ABX with plain -V0 (which is not a proof for improvement as I did not compare the two mp3 versions, and did not do these ABX tests at the same time. But it was enough to make me happy).

I'm not sensitive to pre-echo, so I'm not the one to tell about audible improvements in this field. But the technical effect of using -V0 together with a) -b 320 and b) a negative value of --ns-xxx applies here as well [a) improves usage of bit reservoir providing more space locally for coding the audio content, b) improves signal-to-noise ratio].

This post has been edited by halb27: Feb 3 2011, 00:37

[2Bdecided]

Samples can be found here.

Oh no they can't!

Cheers,
David.

[markanini]

Anyone tested halb27s custom tunings? Also I wonder what a mere --lowpass 16 would do for -V2 to -V0 . I certainly can't hear anything much above 16KHz at sane levels, of course everyones ears are different.

[/mnt]

Samples of the tracks I used can now be found here.

Also I did a test with halb27's custom tunings:

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.2
2011/02/12 23:02:47

File A: I:\Music\Kraftwerk - The Man Machine\01. The Robots.flac
File B: C:\Temp\Man Machine Test 2011\The Robots halb27.mp3

23:02:47 : Test started.
23:03:15 : 01/01 50.0%
23:03:28 : 02/02 25.0%
23:03:36 : 03/03 12.5%
23:03:50 : 04/04 6.3%
23:04:09 : 05/05 3.1%
23:04:23 : 06/06 1.6%
23:04:39 : 07/07 0.8%
23:04:56 : 08/08 0.4%
23:05:19 : 09/09 0.2%
23:05:48 : 10/10 0.1%
23:05:58 : 11/11 0.0%
23:06:13 : 12/12 0.0%
23:06:15 : Test finished.

----------
Total: 12/12 (0.0%)

Tons of smearing softer pre-echo artifacts, far from transparent though.

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.2
2011/02/12 23:25:10

File A: I:\Music\Kraftwerk - The Man Machine\06. The Man Machine.flac
File B: C:\Temp\Man Machine Test 2011\The Man Machine halb27.mp3

23:25:10 : Test started.
23:25:32 : 01/01 50.0%
23:25:35 : 02/02 25.0%
23:25:39 : 03/03 12.5%
23:25:46 : 04/04 6.3%
23:25:51 : 05/05 3.1%
23:25:56 : 06/06 1.6%
23:26:01 : 07/07 0.8%
23:26:05 : 08/08 0.4%
23:26:15 : 09/09 0.2%
23:26:22 : 10/10 0.1%
23:26:27 : 11/11 0.0%
23:26:33 : 12/12 0.0%
23:26:41 : 13/13 0.0%
23:26:47 : 14/14 0.0%
23:26:51 : 15/15 0.0%
23:26:52 : Test finished.

----------
Total: 15/15 (0.0%)

Harsh pre-echo, far away from transparent.

320 CBR vs halb27 320

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.2
2011/02/13 15:57:34

File A: C:\Temp\Man Machine Test 2011\Metropolis 320 CBR.mp3
File B: C:\Temp\Man Machine Test 2011\Metropolis halb27.mp3

15:57:34 : Test started.
15:58:12 : 01/01 50.0%
15:58:15 : 02/02 25.0%
15:58:19 : 03/03 12.5%
15:58:22 : 04/04 6.3%
15:58:25 : 05/05 3.1%
15:58:29 : 06/06 1.6%
15:58:33 : 07/07 0.8%
15:58:36 : 08/08 0.4%
15:58:40 : 09/09 0.2%
15:58:43 : 10/10 0.1%
15:58:46 : 11/11 0.0%
15:58:51 : 12/12 0.0%
15:58:52 : Test finished.

----------
Total: 12/12 (0.0%)

halb27's mod tunings sound better then the default 320 CBR encoding.

I'm curious how these tracks do with Ogg Vorbis. This is one of my all time favorite albums! Maybe its time to stop lurking and start ABXing...

Ogg Vorbis is well known to struggle at the start on The Robots track:

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.2
2011/02/12 22:55:22

File A: I:\Music\Kraftwerk - The Man Machine\01. The Robots.flac
File B: C:\Temp\Man Machine Test 2011\The Robots q 6.ogg

22:55:22 : Test started.
22:55:43 : 01/01 50.0%
22:55:58 : 02/02 25.0%
22:56:07 : 03/03 12.5%
22:56:24 : 04/04 6.3%
22:56:36 : 05/05 3.1%
22:56:48 : 06/06 1.6%
22:57:07 : 07/07 0.8%
22:57:17 : 08/08 0.4%
22:57:36 : 09/09 0.2%
22:57:43 : 10/10 0.1%
22:57:50 : 11/11 0.0%
22:57:58 : 12/12 0.0%
22:57:59 : Test finished.

----------
Total: 12/12 (0.0%)

The synth at 0:07 and 0:15, gets smeared really badly. I used oggenc2 (Xiph.Org libVorbis I 20101101 (Schaufenugget)).

Also there is some slight smearing on the last track with Ogg Vorbis:

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.2
2011/02/12 23:20:51

File A: I:\Music\Kraftwerk - The Man Machine\06. The Man Machine.flac
File B: C:\Temp\Man Machine Test 2011\The Man Machine q 6.ogg

23:20:51 : Test started.
23:21:34 : 01/01 50.0%
23:21:43 : 02/02 25.0%
23:21:55 : 03/03 12.5%
23:22:11 : 03/04 31.3%
23:22:18 : 03/05 50.0%
23:22:21 : 04/06 34.4%
23:22:25 : 05/07 22.7%
23:22:32 : 06/08 14.5%
23:22:36 : 07/09 9.0%
23:22:44 : 08/10 5.5%
23:22:53 : 09/11 3.3%
23:23:09 : 10/12 1.9%
23:23:31 : 11/13 1.1%
23:23:39 : 12/14 0.6%
23:23:44 : 12/15 1.8%
23:23:49 : 13/16 1.1%
23:23:53 : 14/17 0.6%
23:23:59 : 15/18 0.4%
23:24:04 : 16/19 0.2%
23:24:10 : 17/20 0.1%
23:24:20 : Test finished.

----------
Total: 17/20 (0.1%)

[halb27]

Thanks for testing my version, /mnt.
Sure there was no hope for transparency, but for improvement. Glad you found it like that.

Just to be sure: did you use my version of Lame, or did you use the options I gave together with the usual Lame version?

This post has been edited by halb27: Feb 13 2011, 18:59

[/mnt]

Just to be sure: did you use my version of Lame, or did you use the options I gave together with the usual Lame version?

I have used your version of LAME, along with your custom options.

[halb27]

Thanks for clarification.

[B7k]

Can their be testing done with aoTuV Beta6.02?

[punkrockdude]

Halb27:
What is the difference beteen your LAME version and the official? Can you also explain what each parameter in your command line is doing or why you use it? Regards.

[halb27]

-b 320 -F
This provides maximum data space for audio data.
mp3 data is a data stream of frame data (container data). Each frame corresponds to a certain number of wave samples (1152 in the case of CD originated music). A frame provides the data space for audio data, and when we usually talk of bitrate this corresponds to the frame bitrate, that is the size of the frame(s). Using CBR frame bitrate is always constant. Using VBR means frame bitrate can vary from frame to frame.
So far this has to do with audio data only with respect to the provided data space for the audio data.
The audio coding of the 1152 samples a frame corresponds to does not necessarily fill up the data space provided by the frame. If necessary the unused space can be used by the following frame(s) data within certain restrictions. This is the concept of bit reservoir.
-b 320 -F provides a stream of 320 kbps frames which is the maximum possible for mp3. In other words data space for coding the audio data is provided to the max. This way those situations are minimized where the encoder runs out of data space with respect to the amount of data the audio mechanism of Lame's VBR demands according to the quality parameter(s) used.
This is all valid for standard Lame, but the probility for situations where the encoder runs out of space is higher when demanding for a better signal-to-noise ratio as I do.
The drawback of -b 320 -F while using VBR is that the resulting file size is the same as that of a CBR 320 encoding though there is a certain amount of unused data space in the frames. Luckily the great lossless mp3packer procedure gets rid of that.

-Y
Takes care that no data space is wasted for the wrong reasons - as long as the sfb21 issue is concerned. Without -Y quality demand for sfb21 (the frequency range beyond 16 kHz) can lead to an increase of audio data for the lower frequencies for no other but technical reasons (missing scale factor for sfb21). That's why -Y is used for -V 3 and below.
The drawback is that accuracy of sfb21 coding can be worth than when not using -Y, but for me this is of minor importance as the frequencies beyond 16 kHz need not be approximated with maximum quality. AFAIK nobody ever complained about -V 3 quality because -Y is used internally.
Like -b 320 -F, -Y has nothing to do with my special Lame version but is plain standard.

--ns-bass -9 --ns-alto -7 --ns-treble -6
These parameters exist already in standard Lame, but this is the place where I deviate a bit from standard Lame. Moreover standard Lame is not totally correct with respect to the usage of --ns-treble and --ns-sfb21.
A negative value for --ns-bass/--ns-alto/--ns-treble increases the signal-to-noise ratio for the bass/medium/high frequencies, thus making Lame more defensive (but increasing the amount of audio data).
With standard Lame the most defensive value you can use is -8. I changed this with my version to -12.
I also changed (increased) the limit frequencies for --ns-bass and --ns-alto. I didn't care whether this still corresponds to bass and medium frequencies but use --ns-bass/--ns-treble/--ns-alto only for deviding things into efficiency classes. The background is that increasing signal-to-noise-ratio for low frequencies is pretty cheap (in terms of increased amount of audio data), but for high frequencies it's expensive. So --ns-bass/--ns-alto/ns-treble allows to differentiate here, and this has still to do with low/medium/high frequencies.
With --ns-treble I leave alone sfb21 in contrary to standard Lame, and --ns-sfb21 does not exist with my version. This is where standard Lame is faulty.
Because --ns-treble is so expensive I gradually reduce the amount of increased signal-to-noise-ratio demands. sfb16 is the first sfb --ns-treble takes care of, and the --ns-treble value applies as given. For sfb20 in contrary there is no more inreased signal-to-noise ratio demand. This gradually reduction of the --ns-treble value does not apply to short blocks which are used when pre-echo problems are encountered.

--noreplaygain
I always used it because I don't need Lame doing replaygain calculations. I use mp3gain for this purpose. Moreover some time ago somebody reported a negative effect when --noreplaygain is not given in the command line. Sorry I don't remember the details.

For those interested in the changed Lame source the changes are here.

This post has been edited by halb27: Mar 2 2011, 01:59

[punkrockdude]

Thank you for the very informative information, Halb27! I will check out your encoder and the command line you use. Regards.

[mjb2006]

--noreplaygain
I always used it because I don't need Lame doing replaygain calculations. I use mp3gain for this purpose. Moreover some time ago somebody reported a negative effect when --noreplaygain is not given in the command line. Sorry I don't remember the details.

If there is a negative effect, we need to mention it in the wiki. But I didn't see any problems with --noreplaygain mentioned in the first 10 pages of threads returned by typing "noreplaygain" in the Google search box on this page. People do say that --noreplaygain makes encoding go faster, and some say that "few, if any" players actually use the ReplayGain info when it's stored in the LAME tag (as opposed to the more common location for storing such info in MP3s: an ID3v2 tag). I don't know if this is still true.

[greynol]

If there is a negative effect, we need to mention it in the wiki.

Don't hold your breath.

[halb27]

As greynol says don't care much about --noreplaygain.
The disadvantage I reported when not using it is probably not related to audio quality because I guess I would have kept it in mind in this case (and I couldn't find the spot either with a quick search). I just remember having read a sentence backing up with some fact that I have used --noreplaygain for a long time. This is really not a good basis for anything.
But as the replaygain information in the Lame tag is also not a good idea it wouldn't be bad to recommend using foobar (or any other tool) to store the replaygain information in those places where any replaygain aware player can use it, or IMO even better, to use mp3gain for changing the mp3stream in such a way that loudness is directly changed (done once this is a lossless process). IMO it's a major advantage of mp3 (and AAC) that the replaygain procedure can be done to the stored audio stream without reencoding thus bringing replaygain to absolutely any player.

This post has been edited by halb27: Mar 3 2011, 10:25

[Donunus]

these songs seem fine with nero aac. I encode at .75 at a minimum though