V5 sample becomes worse in 3.99, compared to 3.98 Options

[Boiled Beans]

I usually play -V5 MP3 files through my computer speakers.

I updated some tags on the original FLAC rip and decided to re-convert the modified FLAC files to MP3 using the latest LAME 3.99.5, instead of just duplicating the tags to the old MP3, which was 3.98.2.

So I could very clearly hear that the new 3.99.5 was worse than 3.98.2 on this sample through my computer speakers. There's more 'wooshy' sound on the cymbals.

Could this be because 3.99 generally uses lower bitrates for V5 compared to 3.98?

Here are three samples
1) the original FLAC
2) the newly converted 3.99.5 sample from the FLAC
3) the old 3.98.2 sample cut using MP3splt. (Didn't re-encode from original FLAC since I don't have old versions of LAME installed)

The song is a live version of Gardening at Night by R.E.M., track 15 on the 1992 I.R.S. Vintage Years reissue of Murmur.

This post has been edited by Boiled Beans: Aug 25 2012, 17:53

Attached File(s)

 15_Gardening_at_Night_Live_EDIT_FLAC.flac ( 3.86MB ) Number of downloads: 86
 15_Gardening_at_Night_Live_EDIT_3.995.mp3 ( 517.3K ) Number of downloads: 109
 15_Gardening_at_Night_Live_EDIT_3.982.mp3 ( 522.23K ) Number of downloads: 94

 

[Kohlrabi]

I bet you have ABX results to back up that claim? I couldn't really hear any obvious flaws, but it'd be easier to compare the files if the samples were aligned, too.

Could this be because 3.99 generally uses lower bitrates for V5 compared to 3.98?

Bitrate is no quality metric when assessing different encoders.

This post has been edited by Kohlrabi: Aug 25 2012, 22:24

[lvqcl]

LAME 3.98.4 encode is attached to the post.

ABXing 3.98.4 from 3.99.5 was easy:

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.14a
2012/08/26 01:34:43

File A: D:\test\15_Gardening_at_Night_Live_EDIT_3.984.mp3
File B: D:\test\15_Gardening_at_Night_Live_EDIT_3.995.mp3

01:34:43 : Test started.
01:35:19 : 01/01  50.0%
01:35:38 : 02/02  25.0%
01:35:42 : 03/03  12.5%
01:35:49 : 04/04  6.3%
01:35:57 : 05/05  3.1%
01:36:01 : 06/06  1.6%
01:36:05 : 07/07  0.8%
01:36:13 : 08/08  0.4%
01:36:17 : Test finished.

----------
Total: 8/8 (0.4%)

Attached File(s)

 15_Gardening_at_Night_Live_EDIT_3.984.mp3 ( 554.74K ) Number of downloads: 95

 

[Kohlrabi]

Oh, well I could have created my own encodes, that was embarrassing, mea culpa.

[Boiled Beans]

Thanks lvqcl for providing a LAME 3.984 encode from the original FLAC.

I didn't do an ABX at first since the cymbal 'whooshiness' was so obvious.

Here are my ABX results using the 3.995 from my post and the 3.984 from lvqcl. I focused on the first few seconds of the song.

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.13
2012/08/26 16:02:26

File A: C:\Downloads\15_Gardening_at_Night_Live_EDIT_3.984.mp3
File B: C:\Downloads\15_Gardening_at_Night_Live_EDIT_3.995.mp3

16:02:26 : Test started.
16:02:44 : 01/01  50.0%
16:02:58 : 02/02  25.0%
16:03:08 : 03/03  12.5%
16:03:17 : 04/04  6.3%
16:03:26 : 05/05  3.1%
16:03:38 : 06/06  1.6%
16:03:49 : 07/07  0.8%
16:04:08 : 08/08  0.4%
16:04:22 : 09/09  0.2%
16:04:26 : Test finished.

----------
Total: 9/9 (0.2%)

This post has been edited by Boiled Beans: Aug 26 2012, 09:08

[lvqcl]

The sample encoded with LAME 3.99.5 -V 4.999 is attached to the post.

Lame 3.98.4 -V 5 => 144 kbps
Lame 3.99.5 -V 5 => 135 kbps
Lame 3.99.5 -V 4.999 => 142 kbps

The bitrate is closer to 3.98.4 -V5, but the quality isn't (ABX result = 8/8).

Attached File(s)

 15_Gardening_at_Night_Live_EDIT_3995_4999.mp3 ( 547.35K ) Number of downloads: 76

 

[IgorC]

3.98 and 3.99 have gone into different direction of development. Tuning of quality at high bitrates.

It will be great if somebody will post the results of 3.97 V5 vbr new, 3.98 V 5.7 and 3.99 V4.99 for, lets say, at least 10 samples. And see whether there was any substantial improvements, if any.

This post has been edited by IgorC: Aug 27 2012, 15:36

[Martel]

This explanation seems strange. If they wanted to improve the higher bitrates, they wouldn't need to screw up anything else.

E.g.

if -V0 then do the "improved" algorithm, else do the original algorithm

[IgorC]

Unless V0 had rate up to 320 to go for difficult samples while V5 had to hit 130 kbps on average.
Shortly, nobody complains that 3.99.5 V0 causes the higher bitrate than 3.98.4/3.97 as it's the highest V setting while V5 should necesary hit 130-135 kbps on average.

[greynol]

I think you mean to say surely instead of shortly.

[Boiled Beans]

Since the R.E.M. sample in the first post, I've been paying attention to my LAME 3.99 V5 encodes recently.
If something sounds bad, I go back to the original rip to compare whether it was bad from the start, and whether encoding with LAME 3.98 would improve over 3.99.

Here's another example, where 3.98.4 is better than 3.99.5.
It's 'Cherish', from Madonna's 'Immaculate Collection'.

I have provided the original FLAC and 3.98.4 and 3.99.5 encodes.

There's more flanging sound (from a tambourine?) in the 3.99.5 version.

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.13
2012/09/08 18:56:45

File A: C:\Downloads\Sound Test\Test\3.995\EDIT 14 Madonna - Cherish 3995.mp3
File B: C:\Downloads\Sound Test\Test\3.984\EDIT 14 Madonna - Cherish 3984.mp3

18:56:45 : Test started.
19:00:24 : 01/01  50.0%
19:01:01 : 02/02  25.0%
19:01:22 : 03/03  12.5%
19:01:44 : 04/04  6.3%
19:02:07 : 05/05  3.1%
19:02:55 : 06/06  1.6%
19:03:27 : 07/07  0.8%
19:03:55 : 08/08  0.4%
19:03:59 : Test finished.

----------
Total: 8/8 (0.4%)

This post has been edited by Boiled Beans: Sep 8 2012, 14:43

Attached File(s)

 EDIT_14_Madonna___Cherish.flac ( 3.52MB ) Number of downloads: 45
 EDIT_14_Madonna___Cherish_3984.mp3 ( 521.83K ) Number of downloads: 87
 EDIT_14_Madonna___Cherish_3995.mp3 ( 500.44K ) Number of downloads: 87

 

[Boiled Beans]

Two very obvious samples are attached in this post.

These tracks are from the Pixies - Complete B Sides, from 2001.

The cymbals are more 'swooshy' on the 3.99.5 samples, as compared to the 3.98.4 samples.

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.13
2012/10/30 22:49:56

File A: C:\EAC Temp\02 Pixies - Vamos (Live) EDIT 3984.mp3
File B: C:\EAC Temp\02 Pixies - Vamos (Live) EDIT 3995.mp3

22:49:56 : Test started.
22:50:14 : 01/01  50.0%
22:50:36 : 02/02  25.0%
22:50:44 : 03/03  12.5%
22:50:51 : 04/04  6.3%
22:51:02 : 05/05  3.1%
22:51:11 : 06/06  1.6%
22:51:19 : 07/07  0.8%
22:51:29 : 08/08  0.4%
22:51:31 : Test finished.

----------
Total: 8/8 (0.4%)

CODE

foo_abx 1.3.4 report
foobar2000 v1.1.13
2012/10/30 22:51:47

File A: C:\EAC Temp\03 Pixies - In Heaven (Live) EDIT 3984.mp3
File B: C:\EAC Temp\03 Pixies - In Heaven (Live) EDIT 3995.mp3

22:51:47 : Test started.
22:53:49 : 01/01  50.0%
22:54:11 : 02/02  25.0%
22:54:22 : 03/03  12.5%
22:54:31 : 04/04  6.3%
22:54:50 : 05/05  3.1%
22:55:01 : 06/06  1.6%
22:55:12 : 07/07  0.8%
22:55:22 : 08/08  0.4%
22:55:23 : Test finished.

----------
Total: 8/8 (0.4%)

This post has been edited by Boiled Beans: Oct 30 2012, 17:44

Attached File(s)

 02_Pixies___Vamos__Live__EDIT.flac ( 3.84MB ) Number of downloads: 37
 02_Pixies___Vamos__Live__EDIT_3984.mp3 ( 509.02K ) Number of downloads: 54
 02_Pixies___Vamos__Live__EDIT_3995.mp3 ( 471.46K ) Number of downloads: 54
 03_Pixies___In_Heaven__Live__EDIT.flac ( 3.24MB ) Number of downloads: 29
 03_Pixies___In_Heaven__Live__EDIT_3984.mp3 ( 510.46K ) Number of downloads: 57
 03_Pixies___In_Heaven__Live__EDIT_3995.mp3 ( 471.41K ) Number of downloads: 222

 

[IgorC]

The cymbals are more 'swooshy' on the 3.99.5 samples, as compared to the 3.98.4 samples.

My findings are quite the same.
3.99.5 preserves better tonal samples while does worse on transients (cymbals in this case).
Roughly speaking 3.98.4 - V5.7 (~135 kbps) should be better for rock music while 3.99.5 -V5 or -V4.99 (~135 kbps) is better for classic music.

P.S. 3.99.5's V5 doesn't yield the same bitrate comparing to 3.98.4's -V5

This post has been edited by IgorC: Oct 31 2012, 03:45

[themanintheshado...]

Music files encoded in LAME V5 don't sound great on anything except portable listening situations[/url] (i.e. earbuds connected to a DAP):

CODE

Portable: listening in noisy conditions, lower bitrate, smaller file size

-V4 (~165 kbps), -V5 (~130 kbps) or -V6 (~115 kbps) are recommended.

-V6 produces an "acceptable" quality, while -V4 should be close to perceptual transparency.

Better results would be achieved by increasing the bitrate:

CODE

Very high quality: HiFi, home, or quiet listening, with best file size

-V0 (~245 kbps), -V1 (~225 kbps), -V2 (~190 kbps) or -V3 (~175 kbps) are recommended.

These VBR settings will normally produce transparent results. Audible differences between these presets may exist, but are rare.
[edit] Very high quality: HiFi, home, or quiet listening, with maximum file size

-b 320 is an alternative to the VBR settings above.

This CBR mode will maximize the MP3's bitrate and overall file size. The extra space may allow for some parts of the audio to be compressed with fewer sacrifices, but to date, no one has produced ABX test results demonstrating that perceived quality is ever better than the highest VBR profiles described above.

RE: Problems using V5 with LAME 3.99.5

And people wonder why some still use LAME 3.97? New isn't always better.