Request for test: Lame cbr 128 Options

[Gabriel]

I have uploaded 3 test versions on gabriel.mp3-tech.org/lame: lameX, lameZ and lame3.

It would be helpfull if those sensitive to high freq problems with 3.96 preset cbr 128 could give me some indications regarding those 3 test versions: do they improve situation against 3.96 when encoding in 128kbps cbr?

[dev0]

Do you want us to compare the three versions to vanilla 3.96 (like the one on your site one?) or is it already included?

[Gabriel]

No, 3.96 is not included in the 3 test versions

[Gabriel]

Now that Roberto's test is over, I will perhaps receive results on this test?

Right now I've got 0 answer in 1 month.

[dev0]

Sorry for the delay. I did a quick test on giveuptheghost-sincealways.sample18sec.flac, which is an obvious problem for lame at this bitrate.

CODE

ABC/HR Version 1.0, 6 May 2004
Testname: lame 128kbps cbr giveuptheghost-sincealways

1L = dec\giveuptheghost-sincealways.sample18sec.3_96.wav
2L = dec\giveuptheghost-sincealways.sample18sec.X.wav
3R = dec\giveuptheghost-sincealways.sample18sec.3.wav
4R = dec\giveuptheghost-sincealways.sample18sec.Z.wav

---------------------------------------
General Comments:
weird mix of pre-echo(?) and rining during the first 10 seconds
---------------------------------------
1L File: dec\giveuptheghost-sincealways.sample18sec.3_96.wav
1L Rating: 2.5
1L Comment: heavy pre-echo
---------------------------------------
2L File: dec\giveuptheghost-sincealways.sample18sec.X.wav
2L Rating: 3.5
2L Comment: hihat rings but is the clearerst of the 4
least pre-echo
---------------------------------------
3R File: dec\giveuptheghost-sincealways.sample18sec.3.wav
3R Rating: 3.0
3R Comment: Obvious 'clicks' within the first seconds
---------------------------------------
4R File: dec\giveuptheghost-sincealways.sample18sec.Z.wav
4R Rating: 3.0
4R Comment:
---------------------------------------
ABX Results:
dec\giveuptheghost-sincealways.sample18sec.3_96.wav vs dec\giveuptheghost-sincealways.sample18sec.X.wav
   9 out of 10, pval = 0.011
dec\giveuptheghost-sincealways.sample18sec.X.wav vs dec\giveuptheghost-sincealways.sample18sec.3.wav
   8 out of 9, pval = 0.020
dec\giveuptheghost-sincealways.sample18sec.3.wav vs dec\giveuptheghost-sincealways.sample18sec.Z.wav
   9 out of 10, pval = 0.011

The score of 3R should have been lower (around 2.7) due to the very annoying clipping it produces. I'll try to verify these results tomorrow and find more problem samples.

[ViPER1313]

Edit - Clip used was 41_30sec.wav

LameX / LameZ are the winners of this test, stock 3.96 worst overall.

ABC/HR Version 0.9b, 30 August 2002
Testname: Lame --preset cbr 128 Test

1R = Z:\Music\Test Samples\lameZ.wav
2R = Z:\Music\Test Samples\lame3.wav
3R = Z:\Music\Test Samples\3.96 Stock.wav
4L = Z:\Music\Test Samples\lameX.wav

---------------------------------------
General Comments:
2 and 3 "skip" at around the 7 second mark while 1 and 4 do not, 2 and 3 also seem to have more warble than 1 and 4. Tested mainly in the 1-3 second range (click was not used to ABX files.)
---------------------------------------
1R File: Z:\Music\Test Samples\lameZ.wav
1R Rating: 4.3
1R Comment: Slight loss of clarity, no real glaring issues, little bright and no pop at 7 seconds - cant ABX vs. 4.
---------------------------------------
2R File: Z:\Music\Test Samples\lame3.wav
2R Rating: 3.5
2R Comment: Warbles, skips, bad overall
---------------------------------------
3R File: Z:\Music\Test Samples\3.96 Stock.wav
3R Rating: 3.0
3R Comment: Warbles like 2, skips, and the tamberines sound extremely harsh compared to the origional and the other samples (evident in 1-2 second range)
---------------------------------------
4L File: Z:\Music\Test Samples\lameX.wav
4L Rating: 4.3
4L Comment: 1 and 4 sound closest to the origional, no pops / clicks, very slight loss of clarity, slight under water sound (evident in all 128kbps recordings)
---------------------------------------
ABX Results:
Z:\Music\Test Samples\lameZ.wav vs Z:\Music\Test Samples\lame3.wav
10 out of 10, pval < 0.001
Z:\Music\Test Samples\lameZ.wav vs Z:\Music\Test Samples\3.96 Stock.wav
10 out of 10, pval < 0.001
Z:\Music\Test Samples\lameZ.wav vs Z:\Music\Test Samples\lameX.wav
1 out of 2, pval = 0.750
Z:\Music\Test Samples\lame3.wav vs Z:\Music\Test Samples\3.96 Stock.wav
10 out of 10, pval < 0.001
Z:\Music\Test Samples\lame3.wav vs Z:\Music\Test Samples\lameX.wav
10 out of 10, pval < 0.001
Z:\Music\Test Samples\3.96 Stock.wav vs Z:\Music\Test Samples\lameX.wav
10 out of 10, pval < 0.001

This post has been edited by ViPER1313: Jun 17 2004, 01:51

[dev0]

What sample did you use?

Please people, help testing. Even if it's just one sample. It doesn't take long.
This is not the HA.org community I used to like/admire so much.

[auldyin]

Would love to help out but my ears, which have been stuck on the side of my head (one on either side), for over 60 years are absolutely gubbed!!

Sorry...................., I can't hear you (echo, echo, echo)


auldyin

[ViPER1313]

What sample did you use?

Please people, help testing. Even if it's just one sample. It doesn't take long.
This is not the HA.org community I used to like/admire so much.

Bah - it was late - my old favorite 41_30sec.wav .

[Gabriel]

Thank you for your help.
I think that I now have enough information to know in wich direction I should dig up.

[[proxima]]

With rebel.wav and applaud.wav samples, lameX is the best according to my tastes. LameZ is the worst.

This post has been edited by [proxima]: Jun 17 2004, 11:44

[Gabriel]

I uploaded 2 new test versions on http://gabriel.mp3-tech.org/lame
lameZs and lameX0. It would be helpfull if you could compare them using cbr 128kbps against 3.96.

No need for extensive tests, just quick tests are enough.

[[proxima]]

A quick test with LisztBMinor.wav and Atom_heart_mother.wav denotes no progression with both versions against v3.96. Ringing/HF issue is always present.
Difficult to say which one is better, all the three are annoying.

[Gabriel]

Thank you. No need for further testing with those ones

[Gabriel]

Once again, new test versions:
lamel159
lameX99
lameZl

Same thing: just a quick test would be fine

[[proxima]]

Unfortunately all the three new experimental versions are not better than 3.96 stable, the samples tested are the same as above. As before, ranking is very difficult.

This post has been edited by [proxima]: Jun 21 2004, 00:12

[Gabriel]

I uploaded new test versions:
lameX1
lame_over1
lame_under1

[wabbit]

1. X1
2. under1
3. 3.96
4. over1
5. 3.90.3

Just a quick 2 minutes test with my favourite sample (from Nightwish - Over the Hills and Far Away).


[Edit]
Amnesia:
1. 3.90.3
2. 3.96
3. X1
4. under1
5. over1 (very bad quality)

[Edit again]
After a small pause, I tried to ABX X1, under1 and 3.96 with the first sample. Failed.
The differences are very small, almost zero. Only over1 is clearly worse.

This post has been edited by wabbit: Jun 24 2004, 21:11

[Atlantis]

Test 1 (almeno_stavolta.wav)

CODE

ABC/HR Version 1.0, 6 May 2004
Testname: gab

1L = E:\Download\test\x1.wav
2L = E:\Download\test\over1.wav
3R = E:\Download\test\under1.wav

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

---------------------------------------
1L File: E:\Download\test\x1.wav
1L Rating: 3.2
1L Comment:
---------------------------------------
2L File: E:\Download\test\over1.wav
2L Rating: 2.8
2L Comment:
---------------------------------------
3R File: E:\Download\test\under1.wav
3R Rating: 4.5
3R Comment:
---------------------------------------
ABX Results:

Test 2 (korn)

CODE

ABC/HR Version 1.0, 6 May 2004
Testname: gab1

1L = E:\Download\test\under1.wav
2R = E:\Download\test\x1.wav
3L = E:\Download\test\over1.wav

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

---------------------------------------
1L File: E:\Download\test\under1.wav
1L Rating: 4.1
1L Comment:
---------------------------------------
2R File: E:\Download\test\x1.wav
2R Rating: 3.6
2R Comment:
---------------------------------------
3L File: E:\Download\test\over1.wav
3L Rating: 4.8
3L Comment:
---------------------------------------
ABX Results:

Best performer seems under1, then over1, then x1.
On a quick test 3.96 is slightly above x1

[Gabriel]

Are you sensitive to high frequency distortions as featured in 3.96?

[Atlantis]

I've tested the 3 versions (under1, over1, x1) and only did a quick comparison with 3.96, cause lack of time.
On the first sample I've abx'd (almeno_stavolta.wav, which I'll upload later) distorsions where easy to pick (both in x1 and 3.96), not so easy with under1, on the second one distortions were much more harder to find.

But, in order to provide more helpful results, I'll test them against 3.96 this weekend.

[[proxima]]

lameX1, lame_over1 and lame_under1 with Atom_hearth_mother.wav sample has still Ringing/HF issue. Difficult to say which one is better. I tested only this sample but i firmly believe that the problem is easy detectable even with others.
Unfortunately this problem seems to be very difficult to resolve

[Gabriel]

Unfortunately this problem seems to be very difficult to resolve

I am starting to think that perhaps what you are hearing could be different from the problem reported by viper.

About the high frequency problem, could it also be in 3.95?

[[proxima]]

I am starting to think that perhaps what you are hearing could be different from the problem reported by viper.

HF problem i'm referring to is the same i noticed with VBR (mostly resolved with --athaa-sensitivity 1). All experimental versions you have provided are affected by this problem. Regarding CBR, during 3.96 beta development you changed quantization selection mode and sfscale for ABR/CBR up to 160kbps because of slight improvement of this damned hf issue (i remember some tests i've done). The problem with 3.96 stable was slightly reduced but is still very annoying for me.

About the high frequency problem, could it also be in 3.95?

Yes, --preset cbr 128 with LAME 3.95.1 is affected too. If necessary (let me know) in these days i will look for the first past version where the problem occurs.

This post has been edited by [proxima]: Jul 7 2004, 14:38

[Gabriel]

Ok, it seems that I mixed 2 problems:
*high frequency problem
*"skipping"