I was testing lame 3.96.1 --preset insane vs. gogo -b 320 using the castanets sample. Here are the results:

Testname: castanets gogo cbr320 vs lame 3.96.1 preset insane

1R = C:\My Test Samples\cast\gogo-3.13-cbr320.mp3.wav
2L = C:\My Test Samples\cast\lame-3.96.1-preset-insane.mp3.wav

---------------------------------------
General Comments:
Sample 1 is clearly better than Sample 2. Sample 2 has very noticeable pre-echo.
---------------------------------------
1R File: C:\My Test Samples\cast\gogo-3.13-cbr320.mp3.wav
1R Rating: 4.0
1R Comment:
---------------------------------------
2L File: C:\My Test Samples\cast\lame-3.96.1-preset-insane.mp3.wav
2L Rating: 3.0
2L Comment:
---------------------------------------
ABX Results:
Original vs C:\My Test Samples\cast\gogo-3.13-cbr320.mp3.wav
10 out of 10, pval < 0.001
Original vs C:\My Test Samples\cast\lame-3.96.1-preset-insane.mp3.wav
10 out of 10, pval < 0.001
C:\My Test Samples\cast\gogo-3.13-cbr320.mp3.wav vs C:\My Test Samples\cast\lame-3.96.1-preset-insane.mp3.wav
9 out of 10, pval = 0.011

IMO Lame 3.96.1 insane preset could use some more tuning as far as this sample is concerned, especially if it's supposed to be the best possible setting quality-wise.

I don't know why gogo cbr320 sounds so much better. Maybe because it uses gpsycho. Or maybe because it uses stereo and not joint stereo by default at 320kbps. I forced joint stereo with gogo just to see the difference, and it sounded similar to lame (annoying). I don't know if this is an isolated case, but I wanted to report this in this forum in case the developers are interested.

edit: addition - I tested lame 3.90.3 --alt-preset insane, and it sounded pretty much the same as 3.96.1 --preset insane. Could someone else verify my results?

Nspsytune is known to be worst at pre-echo handling than GPsycho.

I think that it will not improve things, but could you try 3.96.1 with -b 320 m s ?

Which version of gogo did you use, and where can I get it from?

OK. I gave that a try. My ears' sensitivity is different tonight, but here is the result output:

ABC/HR Version 0.9b, 30 August 2002
Testname: castanets gogo cbr320 vs lame 3.96.1 -b320 -ms

1L = C:\My Test Samples\cast\lame-3.96.1-b-320-ms.mp3.wav
2R = C:\My Test Samples\cast\gogo-3.13-b320.mp3.wav

---------------------------------------
General Comments:
Sample 2 is very good. Sample 1 has a fairly noticeable pre-echo.
---------------------------------------
1L File: C:\My Test Samples\cast\lame-3.96.1-b-320-ms.mp3.wav
1L Rating: 4.0
1L Comment:
---------------------------------------
2R File: C:\My Test Samples\cast\gogo-3.13-b320.mp3.wav
2R Rating: 4.8
2R Comment: This one is very good.
---------------------------------------
ABX Results:
Original vs C:\My Test Samples\cast\lame-3.96.1-b-320-ms.mp3.wav
10 out of 10, pval < 0.001
Original vs C:\My Test Samples\cast\gogo-3.13-b320.mp3.wav
8 out of 10, pval = 0.055
C:\My Test Samples\cast\lame-3.96.1-b-320-ms.mp3.wav vs C:\My Test Samples\cast\gogo-3.13-b320.mp3.wav
10 out of 10, pval < 0.001

I failed to abx orig vs sample 2 (5.5%) this time. By switching to stereo, I think lame showed a little improvement, but gogo still sounded much bette to me. Also note that gogo uses -q 5 for cbr320 by default. I tried to compare -q 5 and -q 0, but they sounded almost identical to me. I somehow felt -q 5 was slightly better .

Anyway, the setting for gogo I used for the above test is simply:

gogo -b 320 castanets.wav gogo-3.13-b320.mp3

Lev: gogo-no-coda ver. 3.13 can be downloaded from here (RareWares).

LoFiYo, would you mind uploading a sample of the GoGo encoded clip? (I'd encode it, but I'm not on Windows). I'd like to look into this myself, after having done some castanets testing with AAC and ATRAC.

so, if gogo cbr 320 sounds better than both lame 3.90.3 and 3.96.1 preset insane, can we interpret this as gogo being superior to preset standard/extreme and even highest quality achievable with lame (preset insane)? that would be a surprising conclusion... i think we'd need to test more samples...

I think that would explain it *on this particular sample*

You answered your own question. More samples should be tested before any conclusions can be made.

You may want to take the following points into account:
- there are far more types of artifacts than just preecho. So, while gpsycho may perform better with preecho-problems, it "could" perform worse in other departments. You cannot know that because the sample used for testing is a preecho-killersample.
- results at 320kbit cannot automatically be extrapolated to --aps and --ape. It could for example be that some artifacts are less noticable at 320kbit, so only preecho remains - while at lower VBR-levels other artifacts come into the game, and we do not know how gpsycho performs on those compared to nspsytune
- there are non-quality-related issues to take into account as well, like for example encoding-speed. To quality-fanatics, such things may be trivial, but to many casual users things like speed isn't unimportant - people are changing encoders and rippers just because of speed-differences(not to say that speed is the main priority - buts it has to be taken into account as well when making decisions)

- Lyx

I will try to upload it in the upload forum.

edit: Done.

My point is not that gogo is superior to lame. It is that lame's preset insane could use more tuning for pre-echo situations.

I was first bothered by lame's weak pre-echo performance with that sample, and used gogo only for comparison. And even though this is probably an isolated case, I thought the developers might be interested.

FYI: gogo is usually about twice or three times as fast as lame, and performs decently. With -q 9, it's about 5-10x as fast, depending on the machine and the settings you use.

what i'm wondering is that with the extra headroom --preset insane offers, would it maybe work better using GPsycho? I'm not suggesting to extrapolate this to --preset standard and/or extreme as well, and it would certainly need listening tests (which I still think would be better spent tuning VBR-new on the fast presets).

LoFi: Yup, i know....... that post wasn't really meant towards you. More just trying to keep some people from panicing "OMG, gogo has better quality than LAME --aps" - which would of course be total nonsense based on the information we have currently.

Short version: more tests needed.

- Lyx

edit: you guys may also want to test first, if this is an isolated case with a killer-sample and if this problem is also noticable with i.e. microattack real-world music (there's a whole genre about that style of music, called "microsound", although i dont like it, but thats a matter of taste