I had interesting results testing Gogo CBR & ABR 160 vs Lame 3.90.3 CBR & ABR 160 vs Radium-FhG ACM/Professional CBR 160 (hq) using the sample fugue.wav. From the FAQ section, I read that ABR is better than CBR, but I guess there are always exceptions to rules. I would like some other people's opinions if they could test gogo cbr 160 on other samples and see if they will get similar results. I will try gogo vs 3.96b1 next.

1L = C:\fugue\fhg-cbr160hq.wav
2R = C:\fugue\gogo-abr160.wav
3R = C:\fugue\3.90.3-alt-preset-cbr160.wav
4L = C:\fugue\3.90.3-alt-preset-abr160.wav
5R = C:\fugue\gogo-cbr160.wav

---------------------------------------
General Comments:
on this sample to me, cbr sounded superior to abr. gogo is
surprisingly good. "gogo -b 160 -q 0 -m j"
---------------------------------------
1L File: C:\fugue\fhg-cbr160hq.wav
1L Rating: 3.5
1L Comment:
---------------------------------------
2R File: C:\fugue\gogo-abr160.wav
2R Rating: 3.0
2R Comment:
---------------------------------------
3R File: C:\fugue\3.90.3-alt-preset-cbr160.wav
3R Rating: 2.2
3R Comment:
---------------------------------------
4L File: C:\fugue\3.90.3-alt-preset-abr160.wav
4L Rating: 2.0
4L Comment:
---------------------------------------
5R File: C:\fugue\gogo-cbr160.wav
5R Rating: 4.0
5R Comment:
---------------------------------------
ABX Results:
Original vs C:\fugue\gogo-cbr160.wav
8 out of 8, pval = 0.004

I just tested lame 3.96b1 also.

I used:
lame396b1 --preset 160
lame396b1 --preset cbr 160
gogo312 -b 160 -m j -q 0


1R = C:\lame3.96b1-preset-abr160.mp3.wav
2R = C:\lame3.96b1-preset-cbr160.mp3.wav
3L = C:\gogo-cbr160-q0.mp3.wav

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

---------------------------------------
1R File: C:\lame3.96b1-preset-abr160.mp3.wav
1R Rating: 3.0
1R Comment:
---------------------------------------
2R File: C:\lame3.96b1-preset-cbr160.mp3.wav
2R Rating: 3.0
2R Comment:
---------------------------------------
3L File: C:\gogo-cbr160-q0.mp3.wav
3L Rating: 3.9
3L Comment:
---------------------------------------

I hope this helps the Lame-developers in one way or another.

About the general comments, you edited the file after saving the results, presumably? Or do they refer to previous listening tests you've done? abchr doesn't allow you to know which codec you're listening to during the test.

ff123

BUSTED!!! (just kidding)

LoFiYo, could you also try to ABX the difference between the different samples? I'm not sure what the official stance is on this at HA, but I'm guessing that, if you can't ABX the difference between two encoded samples, then you can't really say that one's better than the other. Of course, in the listening tests given by Roberto, no one's expected to do 60 zillion ABX tests, but for this one instance, maybe it wouldn't be too much trouble to run a couple Sample vs. Sample tests?

I would also suggest you upload this sample you used.

Additionally to what ff123 said , you might want to ABX the codecs against each other in ABX/HR session (5/5 should be enough) to make sure there's really a difference. If you want to discuss your results publicly here, a link to the sample you used (-> upload?) and some description about the artifacts you heard would make sense. Results like yours are appreciated and necessary to find out where the actual lame version stands compared to the currently recommended version 3.90.3, but without these details your result isn't a big help for this.

Thanks everybody for your replies. I will try to upload the sample later this week (if I can figure out how!). Actually I think it's pretty famous and publicly available somewhere (I've seen several regulars here mention it in the past - e.g. 2BDECIDED, Guruboolez etc.). I got it long ago from a HA thread where Guruboolez used it in his post, but now I can't find the link to it.

ff123, Yes, I added the comment section after the test was performed and the result text file was saved, which I now realize is a no-no. I also changed the filenames so that people can tell what encoders were used in what settings etc. But the test was performed totally blind (it's impossible to cheat your program ). Now I think that I should not have done that and left the result test file untouched.

Also I guess I should have taken more time and ABX'd each encodes against one another. I was just so surprised to see GOGO CBR 160 perform so well with this sample despite the popular beliefs (ABR > CBR; LAME > GOGO).

iirc gogo uses GPSYCHO only. You may want to compare 3.90.x against gogo at the same playground too.

So are you suspecting that nspsytune may be the culprit of the apparently lower performance of the Lame presets on this sample?

As I promised, I did a more complete ABX test. This time I didn't f*** around with the result text file. This shows that I wasn't delusional the first time.

Haven't had a chance to figure out how to upload the sample yet.

ABC/HR Version 0.9b, 30 August 2002
Testname: FUGUE GOGO CBR160 VS LAME ABR/CBR 160

1L = C:\My Music\lab\fugue\lame396b1-preset-cbr-160.mp3.wav
2L = C:\My Music\lab\fugue\gogo312-160-q0.mp3.wav
3R = C:\My Music\lab\fugue\lame396b1-preset-abr-160.mp3.wav

---------------------------------------
General Comments:
SAMPLE 2 SOUNDED THE BEST. 1 & 2 SOUNDED PRETTY MUCH THE SAME. 3 WAS MAYBE A LITTLE BETTER.

ABX MP3 & MP3 WAS OF COUSE TOUGHER. 1 VS 2 WAS EASIER (STILL TOUGH) THAN 2 VS 3. 1 VS 3 WAS IMPOSSIBLE. BY THE TIME I GOT TO 1 VS 3, MY HEAD WAS SPINNING.
---------------------------------------
1L File: C:\My Music\lab\fugue\lame396b1-preset-cbr-160.mp3.wav
1L Rating: 2.9
1L Comment: FIRST CODE IS SMEARED. DIRTY SOUNDING. VERY EASY TO ABX AGAINST THE ORIGINAL.
---------------------------------------
2L File: C:\My Music\lab\fugue\gogo312-160-q0.mp3.wav
2L Rating: 3.6
2L Comment: THIS ONE WAS THE HARDEST TO ABX AGAINST THE ORIGINAL. HAD TO LISTEN UP TO 10+ TIMES EACH TIME TO GET IT RIGHT.
THE FIRST CODE/HIT SOUNDS MUCH CLEANER THAN 1 OR 2.
---------------------------------------
3R File: C:\My Music\lab\fugue\lame396b1-preset-abr-160.mp3.wav
3R Rating: 3.0
3R Comment: VERY SLIGHTLY CLEANER THAN 1 MAYBE. VERY EASY TO ABX AGAINST THE ORIGINAL.
---------------------------------------
ABX Results:
Original vs C:\My Music\lab\fugue\lame396b1-preset-cbr-160.mp3.wav
8 out of 8, pval = 0.004
Original vs C:\My Music\lab\fugue\gogo312-160-q0.mp3.wav
8 out of 8, pval = 0.004
Original vs C:\My Music\lab\fugue\lame396b1-preset-abr-160.mp3.wav
8 out of 8, pval = 0.004
C:\My Music\lab\fugue\lame396b1-preset-cbr-160.mp3.wav vs C:\My Music\lab\fugue\gogo312-160-q0.mp3.wav
12 out of 14, pval = 0.006
C:\My Music\lab\fugue\lame396b1-preset-cbr-160.mp3.wav vs C:\My Music\lab\fugue\lame396b1-preset-abr-160.mp3.wav
7 out of 17, pval = 0.834
C:\My Music\lab\fugue\gogo312-160-q0.mp3.wav vs C:\My Music\lab\fugue\lame396b1-preset-abr-160.mp3.wav
14 out of 24, pval = 0.271

Hmmm, interesting results. At least LAME CBR didn't beat LAME ABR this time .

How do I do that? Do only admins get to upload?

Edit: Never mind! I figured it out. Please see http://www.hydrogenaudio.org/forums/index....ndpost&p=195031

Hey guys, I did what you told me to do. I submitted more complete test results and the original wav file. Now what? "Interesting" -- that's it?

It would really be interesting to see how you rate lame.3.90.3 "-b 160 -q0 -mj" and "--abr 160 -q0 -mj" against fhg and gogo.

OK. I will try to find time to do that. Thanks for the suggestion

But since the original wav file is publicly available (see above), YOU or anybody who is reading this could do the same! and maybe report what you/they find here also. I would like to hear back something other than "interesting" or "that's nice" if I am to spend some of my spare time doing something suggested here.

It is interesting but so far it is only an isolated case and those are allways expected. The next step would be to test a few more samples and see if they behave the same.

I am suspecting that trumpet.wav would act in the same way.

OK. I tried your suggestion, and found out that your simple command line settings sounded better to me on this sample than the --alt-presets (3.90.3). The -q0 might be the reason for this improvement (the --alt-presets internally use -q2 if I remember correctly). Haven't tried 3.96b1 yet. The results may be inaccurate, because my ears may have been less sentitive today than the other day. I wish someone could verify that. Have you tried it yourself, Robert?

ABC/HR Version 0.9b, 30 August 2002
Testname: fugue - gogo312 cbr160 vs lame 3.90.3 cbr/abr 160

1L = C:\My Music\lab\fugue\3903-abr160-q0-mj.wav
2R = C:\My Music\lab\fugue\3903-b160-q0-mj.wav
3L = C:\My Music\lab\fugue\gogo312-b160-q0.wav

---------------------------------------
General Comments:
failed ABXing between samples With very small differences, the quality of the samples are pretty much the same.
Able to ABX orig vs mp3s. Robert's suggested command line sounds better with this sample than --alt-presets.
---------------------------------------
1L File: C:\My Music\lab\fugue\3903-abr160-q0-mj.wav
1L Rating: 3.9
1L Comment:
---------------------------------------
2R File: C:\My Music\lab\fugue\3903-b160-q0-mj.wav
2R Rating: 3.5
2R Comment:
---------------------------------------
3L File: C:\My Music\lab\fugue\gogo312-b160-q0.wav
3L Rating: 3.7
3L Comment:
---------------------------------------
ABX Results:
Original vs C:\My Music\lab\fugue\3903-abr160-q0-mj.wav
7 out of 7, pval = 0.008
Original vs C:\My Music\lab\fugue\3903-b160-q0-mj.wav
7 out of 7, pval = 0.008
Original vs C:\My Music\lab\fugue\gogo312-b160-q0.wav
7 out of 7, pval = 0.008
C:\My Music\lab\fugue\3903-abr160-q0-mj.wav vs C:\My Music\lab\fugue\3903-b160-q0-mj.wav
5 out of 9, pval = 0.500
C:\My Music\lab\fugue\3903-abr160-q0-mj.wav vs C:\My Music\lab\fugue\gogo312-b160-q0.wav
6 out of 9, pval = 0.254
C:\My Music\lab\fugue\3903-b160-q0-mj.wav vs C:\My Music\lab\fugue\gogo312-b160-q0.wav
7 out of 13, pval = 0.500

I thought -q0 was experimental and not recommended. Shouldn't --alt-preset 160 and --alt-preset cbr 160 have been used?

In my earlier test, I did test --alt-preset 160 and --alt-preset cbr 160 and found them inferior to GOGO CBR 160 on this sample (please see above). I think Robert wanted to see if q0 would help. Well, my last test results apparently indicate that Takehiro's then-experimental -q0 does improve the quality over the recommended -q2 on this sample.

See for yourself

it's not the -q0 making the difference, but the psymodel 3.90.3 defaults to on simple command lines. I was just repeating your gogo command line. To answer your question: "when gogo beats lame..."

then you found one sample where GPSYCHO acts better than NSPSY.

btw later versions of LAME don't give you access to GPSYCHO on beta and stable releases anymore.

Dude you have cracked it GoGo is the answer. Guys! Cancel LAME development! Finally transparency can be produced by some Toyota cut down version of an old LAME build!

Sorry I probably breached a couple of the rules there. What I meant to say was.... "whatever?"

More beer vicar?
Sure

Dude chill out. HA has to use all data that comes if it wants to be considered scientifically minded.

edit: removed excess punctuation (I'm high)

soz i was drunk!