Hi all,

I've always thought musepack had a bass tune bug but because I'm not big on ABXing I never followed this up, until now, here is a sample I think proves this. As you can see from my ABX results I keep to trials of 5 because I lose concentration.

1st

2nd

3rd

4th

5th

i can't hear a difference. perhaps it's encoder clipping? i encoded my sample with --xlevel. will try without it, but if there's a difference without --xlevel i suggest you use that switch.

[edit]

hmm... xlevel and regular are bit identical.

i heard a very very very very faint crunch in one sample that wasn't in the other. perhaps this is it. i can't statistically say i heard it as i lost concentration and couldn't ABX.

[edit edit]

AAARGH!! my sound card's JUST TOO NOISY! dammit.

well, this sample doesn't bother me. but if you really can hear a difference, i must commend your hearing (and equipment). hehe. let's see a 16/16 ABX to make sure it wasn't a fluke

As I said I begin to I lose concentration but I'm sure this is an issue with musepack, I'll continue ABXing when I get free time & try to present better ABX results, in the mean time I have posted the sample to see if other people can also hear this & post their ABX results.

yeah, i read that. still an ABX is probably required for this to be recognised as a problem sample (you know how nazi it can get here ).

it's possible that q6 will solve the issue. but thanks for the sample anyway.

If these are all your ABX-tests, then the result is highly significant:
25 out of 31, p-val = 0.00043 (!)

I decided to give 16 trials a go as MugFunky suggested,



I would like to mention that I found this easier to spot using my Panasonic micro system with speakers than my Sennheiser HD 495 headphones.

Also keep in mind when ABXing it's not an artifact to spot, it's a difference in the bass, the bass is sharper in the original.

I guess I'm blessed (?) with bad hearing... Can't even abx it at -q4!
I'll just go ahead and blame it on noisy onboard sound. Maybe when I get back to the US I'll get a good sound card.

WinABX v0.4 test report
02/17/2004 21:01:15

21:33:45 1/1 p=50.0%
21:34:10 2/2 p=25.0%
21:38:17 3/3 p=12.5%
21:38:43 3/4 p=31.2%
21:38:56 4/5 p=18.8%
21:39:14 5/6 p=10.9%
21:39:48 5/7 p=22.7%
21:40:04 6/8 p=14.5%
21:41:00 7/9 p=9.0%

There is some kind of attenuation in the bass. And yes, the Sennheiser 495S are not good for this, but I can hear it well. I will try with --extreme later and a few other codecs. This should be put on the download section. It's a very valid sample.

One thought about this: Could it be that the difference depends on equipment? The strong base could cause audible distortions of inaudible (masked) differences. Here (Aureon Sky + Sennheiser HD 540) I can't even immagine a difference.

The HD495s don't have enough detail at the very low end. Sadly, I need to disable all ambient noise around me and concentrate. This is the latest Nero AAC encoder at "Standard":

23:02:06 1/1 p=50.0%
23:02:15 2/2 p=25.0%
23:02:29 3/3 p=12.5%
23:02:46 4/4 p=6.2%
23:03:14 4/5 p=18.8%
23:03:30 5/6 p=10.9%

Vorbis GT3b2, -q 6 (a little easier than the other 2):

WinABX v0.42 test report
02/17/2004 23:17:59

A file: D:\hard_knock.wav
B file: D:\hard_knock-vorbis-6.wav

Start position 00:00.0, end position 00:00.5
23:29:15 1/1 p=50.0%
23:29:23 2/2 p=25.0%
23:29:31 3/3 p=12.5%
23:29:43 4/4 p=6.2%
23:29:56 5/5 p=3.1%

I've tried this sample and I can't detect anything unusual at q5.

My first & only (so far) try at quality 10



Bass difference seems to me just as audible as before, I will try again later.

I can hear no difference

It's nearly transparent to me.

foo_abx v1.2 report
foobar2000 v0.8.2
2004/07/03 01:20:01

File A: file://H:\temp\abx tests\hard_knock.flac
File B: file://H:\temp\abx tests\hard_knock.mpc

01:20:01 : Test started.
01:20:19 : 01/01 50.0%
01:20:34 : 02/02 25.0%
01:20:41 : 03/03 12.5%
01:20:48 : 04/04 6.3%
01:21:00 : 05/05 3.1%
01:21:22 : 05/06 10.9%
01:21:37 : 06/07 6.3%
01:21:43 : 06/08 14.5%
01:21:49 : 07/09 9.0%
01:21:55 : 08/10 5.5%
01:22:12 : 09/11 3.3%
01:22:35 : Test finished.

----------
Total: 9/11 (3.3%)


Koss portapro, TB santa cruz.

The mpc file bass seems more pronounced. This is very subtle and I doubt that anyone will notice it in normal listening, I couldn't even abx this one months ago.

Well I'm not big on finding test samples, ABXing but this really, really annoyed me & I heard this kind of thing on other songs, people who like rap music will find this a very annoying artifact.

I believe the start of Metallica-Enter Sandman has this kind of artifact but I don't have an original copy of the CD or think I really need to prove this further after the results of other users.

Often these 'artifacts' are in the original sample as well - I am sure this is an isolated case. Its annoying to you because in this case you proved it and now you will suspect other similar sounding samples to fail.

I try now and again to randomly abx potentialy 'suspect' samples on q5 and fail.