We test a lot of music compressions and have one result:

Lame 3.96.1 cause a bug: The bass on the left side has more power than on the right side.

What do you say about this?

We think Lame 3.90.2 and 3.90.3 still rules...

ABX results?

Sample and abx results please?

both versions using --alt-preset standard

Nas - Last Real ***** Alive

Lame 3.96.1 @ 168 kbps

Lame 3.90.3 @ 190 kbps


using other presets like cbr mode or the DLL Version doesn't remove the problem.

tracks like Nas - Life Is What You Make It and Dr. Knockboot

tested on Philips HP 890 Earphones. We're 5 persons, we all agree with the problem of Lame 3.96.1

Lame 3.90.3 needs mor bits, but there are no bass problems

Then there shouldn't be a problem providing 5 ABX results.

so what do YOU say? Note: We don't have only one Earphone, we have 4 ones.

The Philips HP890 was winner in several magazines which have tested earphones.

http://www.hydrogenaudio.org/forums/index....974#entry149481
http://www.hydrogenaudio.org/forums/index....showtopic=16295

memokai. The forum TOS #8, state you to backup you sound quality statement with a blind listening test (ABX test).
That should not be a problem to perform, since 5 persons clearly hear the difference...

Please read this before posting such comments

Welcome to HA.
I think you mean well.

Oh, and if you are interested in testing the quality of the newer versions of LAME versus the older ones, why don't you use the latest 3.97 alpha?
It's "preset standard" could be better than 3.90.2/3 (still testing myself)

EDIT: Damn! TWO members beat me to it!

Well, not that it's a bad thing to always quote the TOS, but.. in this particular case I think a simple measurement may be enough

Let me explain..

If a consistent volume bias were introduced by LAME, and if it's an obvious bug (ie: not a psychoacoustic effect), then it should also be measurable by a machine..

(don't forget the lowpass, since he said the amplification was on bass only).

1 - Take your preferred <=30 sec. music sample.
2 - Measure the left and right-channel RMS values on a lowpassed copy of the original sample.
3 - Do an encode/decode cycle
4 - Measure the left and right-channel RMS values on a lowpassed copy of the decoded sample.

If the RMS power ratio (right RMS divided by left RMS) has changed significantly after the encoding, then you most probably have a buggy codec.

If, then, the measurements show nothing, well of course, it could be some subtle psychoacoustic issue and an ABX test is needed

Well, we HAVE done some ABX tests, I say how we proceeded:

Test Track: Nas - Life Is What You Make It

A: Encoded with Lame 3.90.2 (--alt-preset standard)
B: Encoded with Lame 3.96.1 (--alt-preset standard)
X: Encoded with Lame 3.90.3 (--alt-preset standard)

Rip-Program: Audiograbber Version 1.83 build 1 (used for all 3 encoders)

4 Persons tested 3 times. They didn't know how the tracks were encoded (I ripped the tracks, made the playlist and then I let them hear the tracks 3 times without looking to the playlist, they had to be "blind")

I said one of the tracks have a different bass on the left side. Say which one it is.

The results:

Person 1: 1st time he couldn't say which one, he guessed B
2nd time: B
3rd time: B

Person 2: 1st: X
2nd: B
3rd: B

Person 3: 1st: B
2nd: B
3rd: B

Person 4: 1st: B
2nd: A
3rd: B

------------> 7 votes for B (Lame 3.96.1) 1 Vote for X (Lame 3.90.3) 1 Vote for A (Lame 3.90.2)

Of course during the test they didn't talk to each other. Beacause then the test would be useless. Then all persons used the same equipment (Philips HP 890)

What do you say? Do you agree when you make such a test?

Shouldn't the original WAV file be included in the ABX tests?

I suggest that you read Pio2001's article to which Latexxx provided a link if you're not familiar with ABX tests.

I'm not sure what you performed was a valid ABX test, but even if it was, the number of trails is too little for the results to be considered relevant.

A valid ABX test usually looks something like this:
- get a software ABX box, such as WinABX or use foobar2000's ABX plugin
- set A to be the original sample
- set B to LAME-encoded sample
- warm up if necessary
- perform 16 trials and see if you can get relevant results

If your results are positive, post the sample so others can verify it. Thanks.

sorry, but I think if Lame 3.90.2 and 3.90.3 don't show the bass bug, why should I compare it with the uncompressed file? The original CD doesn't have the problem, we heard it several times on WAV.

But if you strictly want to have an ABX test, you have to do it on your own. I'm new here, that's for sure, but when you don't believe me, why don't you test it on your own?

I only made the suggestion that there could be a bass-problem. When you don*t agree with the bug, then make some examples, and we're done with it. I have two 3.96.1 Versions (183 kb and one with 197 kb size), downloaded from rarewares.org and mp3dev.org.

We had the idea of the problem since lame 3.96.1 released.

So I ask YOU WHAT DOU THINK ABOUT IT? Do you have the same opinion that there could be a bass problem or not?

greeting memomai, not memokai (ääh)

You have to compare it the WAV file because the whole point is to determine which encoder most closely represents the original. The test is meaningless without that.

Also, the burdon of ABX testing and proving the claim belongs to the person who is making the claims, not us.

Sorry, be we can't just take your word for it, and since you made the claim it's your responsibilty to prove it.

if we want to do the abx test ourself, we need the small sample (10 to 20 sec) of the original wav. Please upload a small excerpt which triggers "the bug".

well, not everybody might be able to hear it. Besides that, do you know how many people make wild claims about things and at the end it turns out that they did something wrong. There are people that claim that one command line sounds better than another one and they'd bet their soul for it...and all of a sudden it turns out, that their setting does the exact same thing than the other one that was supposed to be worse, because it's mapped internally to the same setting...

Out of 100 claims we have here, there's probably one that is valid. So, you either post a valid ABX test, or nobody will believe you...since it is so easy for you to spot, it won't take long...maybe 5 minutes?! So just compare the setting were that problem occured against the original CD file...do it 15 times or so and maybe choose another song and repeat. The sample can be 30 seconds or shorter...