Lame 3.96 (test version) pre-echo test Options

[Gabriel]

http://gabriel.mp3-tech.org/lame/lame_test.exe

This is a test version (not the same one as in cvs) for pro-echo. It uses less short blocks than 3.95, but with a stronger pre-echo control.

I would like reports of this one against 3.95 concerning mainly sharpness.

edit: preset standard please

This post has been edited by Gabriel: Jan 19 2004, 17:19

[guruboolez]

I'll test it, though it isn't the good week (Roberto's listening test).
But I won't post results before thursday or friday...

[tigre]

What settings do you want people to test? Is --preset standard ok or others too?

[tigre]

Terratec Aureon Sky (Envy24ht)
foobar2000 Kernel streaming 24bit padded to 32bit output
replaygain (trackgain, default level) enabled

Quitate.wav sample from
Sergent Garcia - Sin Fronteras (11) Quitate la Arena

ABXed 0.0 - 2.3 with foobar2000 ABX tool

1. Quitate.wav vs. Quitate_3.95.1_aps.mp3 8/8
Comment: Focused on clave (percussion), hits less sharp, smeared

2. Quitate.wav vs. Quitate_test_aps.mp3 8/8
Comment: Similar to 1.

3. Quitate.wav vs. Quitate_3.90.3_aps.mp3 8/8
Comment: Similar to 1.

4. Quitate_3.95.1_aps.mp3 vs. Quitate_test_aps.mp3
Comment: Hard to abx. I got 4/4, then lost the difference. If I have more time to relax between trials, I'll probably be able to do it. The test version seems to be better, but the improvement is small, compared to the difference to original.

5. Quitate_3.90.3_aps.mp3 vs. Quitate_3.95.1_aps.mp3 7/7
Comment: The difference is a little bit bigger than in 4., it seems like 3.95.1 has less pre-echo, but I'll have to re-check this using ABC/HR.

I don't have time to keep on testing right now, so I'll post the results + the sample and probably add more results later.

Attached File(s)

 quitate.flac ( 1.92MB ) Number of downloads: 235

 

[Gabriel]

Did you used the test version? (you did not mention it)

edit: sorry, I did not noticed that it was mentioned in the name of your mp3 files.

This post has been edited by Gabriel: Jan 21 2004, 12:37

[tigre]

ABC/HR results for the same sample:

ABC/HR Version 0.9b, 30 August 2002
Testname: lame preecho - sample: quitate.wav

1L = N:\quitate_test_aps.wav
2R = N:\quitate_3.95.1_aps.wav
3R = N:\Quitate.3.90.3_aps.wav

---------------------------------------
General Comments:
for ABX used 0.0-1.5.
For differences see rankings (sample 3 = 3.0 etc.)
---------------------------------------
1L File: N:\quitate_test_aps.wav
1L Rating: 4.0
1L Comment: preecho
---------------------------------------
2R File: N:\quitate_3.95.1_aps.wav
2R Rating: 3.6
2R Comment: preecho
---------------------------------------
3R File: N:\Quitate.3.90.3_aps.wav
3R Rating: 3.0
3R Comment: preecho
---------------------------------------
ABX Results:
Original vs N:\quitate_test_aps.wav
    8 out of 8, pval = 0.004
Original vs N:\quitate_3.95.1_aps.wav
    8 out of 8, pval = 0.004
Original vs N:\Quitate.3.90.3_aps.wav
    8 out of 8, pval = 0.004
N:\quitate_test_aps.wav vs N:\quitate_3.95.1_aps.wav
    13 out of 15, pval = 0.004
N:\quitate_test_aps.wav vs N:\Quitate.3.90.3_aps.wav
    8 out of 8, pval = 0.004
N:\quitate_3.95.1_aps.wav vs N:\Quitate.3.90.3_aps.wav
    8 out of 8, pval = 0.004

Note that ABXing the mp3 samples against each other was quite hard. The ratings 3.0 for 3.90.3 and 4.0 for 3.95.1 exaggerate the differences between them. The difference between the best mp3 and the original is still (I'd say 3 times) bigger than between the mp3s. I think it could be even possible, that other listeners rate the mp3s in another order.

[Lev]

A big thanks is in order, definately, to everyone involved in testing, and of course, Gabriel and others for developing

A question I have is: Do the tweaks of --aps remain the same between 3.90.3, 3.95 and this test version? And are those tweaks "Optimum fine tunings for the MP3 codec", or will they themselves need tweaking given a new lame, with a modified psymodel / pre-echo control / whatever?

I wish I had some spare time to help testing (I dont even have time at home to even read the forum, really!)

[tigre]

Another test with the same sample, but ABXed at a different position (= different instrument causing pre-echo).

ABC/HR Version 0.9b, 30 August 2002
Testname: lame pre-echo - sample: quitate.wav

1R = N:\Quitate.3.90.3_aps.wav
2L = N:\quitate_3.95.1_aps.wav
3R = N:\quitate_test_aps.wav

---------------------------------------
General Comments:
Position used for ABXing: 6.7-8.2
pre-echo on bells on the right channel
1 is best, abxing vs 2 and 3 is very hard, only possible with breaks after a few trials. abxing 2 vs. 3 is impossible for me with reasonable effort - if there's any difference between the two I couldn't say which is better anyway.
---------------------------------------
1R File: N:\Quitate.3.90.3_aps.wav
1R Rating: 3.5
1R Comment:
---------------------------------------
2L File: N:\quitate_3.95.1_aps.wav
2L Rating: 3.0
2L Comment:
---------------------------------------
3R File: N:\quitate_test_aps.wav
3R Rating: 3.0
3R Comment:
---------------------------------------
ABX Results:
Original vs N:\Quitate.3.90.3_aps.wav
    8 out of 8, pval = 0.004
Original vs N:\quitate_3.95.1_aps.wav
    8 out of 8, pval = 0.004
Original vs N:\quitate_test_aps.wav
    8 out of 8, pval = 0.004
N:\Quitate.3.90.3_aps.wav vs N:\quitate_3.95.1_aps.wav
    19 out of 25, pval = 0.007
N:\Quitate.3.90.3_aps.wav vs N:\quitate_test_aps.wav
    17 out of 22, pval = 0.008
N:\quitate_3.95.1_aps.wav vs N:\quitate_test_aps.wav
    16 out of 27, pval = 0.221

[Gabriel]

A question I have is: Do the tweaks of --aps remain the same between 3.90.3, 3.95 and this test version?

The tuning is different between 3.90 and 3.94. This test version is featuring some slight adjustments to 3.94/3.95.

[guruboolez]

I've tried to ABX 3.95.1 and this 3.96 alpha on some "short blocks samples". Both encoders are really close. I failed on most files (but I had to admit that my ears were tired after Roberto's test and an additional session).
Nevertheless, I've ABXed with success three files :
bachpsichord, and two others (two percussive instruments) I've can't post for the moment. 3.96.a was slightly more affected by pre-echo, but difference was hard to find (for the last file, I've performed three ABX tests before succeeding).

P.S. On these short samples, 3.96a encodings were bigger than 3.95.1 (but more short-blocks for 3.95.1, according to encspot).