See this lnk:

http://forums.syntrillium.com/topic.asp?TO...735&whichpage=1

especially pages 3 and 4. Was I too rash? Does anyone have the CD to check?

Cheers,
David.

I didn't notice there anyone claiming (s)he ABXed MusePack...

Please add a quote.

EDIT: Okay, I've noticed it...

Does that 0.4% means 0.004?
But we'll require him to do ths test again and post the sample
(0.05 is the lowest acceptable probability, anything else might be luck)
Maybe he could use ABX/HR and post whole results?

(Probability is hard to forge... very precise)

teflon20 says that he will try to pass the challenge and then posts a WinABX picture.

Without the samples or more info it's hard to tell anything though.

Might be that we have a new problem sample here, but it's more likely that he's talking out of his arse. This is a quote from him, page 2:
"I've heard the new mp3 codec version in CEP 2.1 at 224kbps and I can say that it redifines the word quality in mp3 codecs. Awesome!" I take it he tried LAME --aps at one point... Yet, he claims he can ABX MPC at q6? Yeah right... Oh wait, he used q5, he says it in his last post. Still.

Another funny quote: "If it's the CD format, big deal. Cause cd ain't high quality." Wrong. It's not about the CD format, it's about how badly it's put to use these days.



Page 4

I think you are confusing percentages and and significance levels. 0.05 = 5%
0.4% = 0.004

The result was highly significant. (He probably ABX'ed till he got a good result instead of deciding a number of trials in advance, but even so it's likely still significant)

I've corrected that before you posted a reply, but my connection is sometimes slow

Anyway, which version of Musepack did he claim to use?
Post-SV7, 1.1, 1.14, 1.15r...

- Several possibilities:

A ) He cheated:

- He used two clearly different wave files to achieve those results.
- He edited the pic.

In the past I thought about adding some kind of secure identification number to the WinABX log report, that took into account the CRC of the wave files and the ABX results. I parked this but I can put it in a future release.


B ) He didn't cheat:

- He didn't use -xlevel and the mpc is clipping.
- He is using a crappy soundcard that makes audible otherwise inaudible things.
- He in fact heard a difference worth of mention.


C ) Something in the middle:

- He run a lot of 16 trials until he got a possitive outcome one.


I think he should explain what command line he used, and put a link to a download location for the song he used. Until someone else duplicates his results, I think it is not safe to say the heard a difference.

Couldn't read on.

That forum is, the exact representation, of why I partake in very few forums except for HA

Lev - I agree - but there's much worse places than the syntrillium forums!

There's
rec.audio.opinion and audio asylum for discussing nothing with any factual basis what-so-ever
the Winamp forums for discussing nothing of any use or consequence what-so-ever
etc etc etc - but I'm sure you've seen them all!

Great to be on HA.

Cheers,
David.

EDIT: He's going to try --quality 6 with a different sound card, and report back. Does anyone here have the Phil Colins Hits Live CD to try? - it can't be that rare!

Actually, from reading the forum there, he said he'd try better headphones, but not that he'd try a different soundcard. With many artifacts, the quality of headphones probably wouldn't matter much so long as they can go loud enough.

As KikeG suggested in his comprehensive list of possible explanations, I suspect (only a guess) that the dodgy resampling on the Soundblaster of his could be revealing something by aliasing a frequency that might be assumed to be masked into another frequency, or that in resampling a particularly sharp transient it has screwed up one of the files (possibly the original) over a broader bandwidth.

Perhaps, David, you could suggest that he resamples both files to 48 kHz using CoolEdit, quality 600 to overcome the limitations of his soundcard's crude hardware resampling and then try to ABX these 48 kHz files.

Sounds like he used mppenc v1.14, thanks to the link at the top of the HA sticky you pointed to.

I guess non-xlevel encoder clipping is also a possible cause that might be hidden by using a front-end that could hide the warning messages. Hopefully when he adds parameters next time he'll use --quality 6 --xlevel as you suggested.

I presume from the blind-testing success of Musepack on HA that decoder-side clipping is always brief enough to be inaudible (though I still use ReplayGain clipping prevention to be sure and only ever tested clipped vs unclipped on one orchestral Lame APS sample that clipped after album mp3gain). Nonetheless, it might really upset the SB soundcard resampler for all I know.

By the way, it's interesting that nobody mentioned Lame (let alone APS) on that forum topic, given its title. The only comments that MP3 could sound good were regarding the internal encoder (FhG?) at 224 kbps and someone also mentioned 256kbps and VBR at around 200 kbps.

If the second ABX comes out positive (even using --quality 5 --xlevel rather than --quality 6) this could be an interesting find (and remarkably lucky, given that he could've chosen practically any CD track in his collection and probably doesn't have a lot of baroque harpsichord CDs ).

P.S. Good to be back on a forum where KikeG's name doesn't get bleeped out! ****G (see p3 of CE forum topic)

I read several threads over there in which mp3 was trashed in general. Many were saying they could hear faults routinely at any bitrate.

Glad to see that some are now trying to prove it anyway.

I have it. I haven't read the forum (off out shortly) but I can rip any time sample from that CD you like.

If the resampling is turning audible a difference, I think it should be taken into account, because most MPC end up playing on resampling soundcards. It would be annoying for an audiophile format not to be able to perform properly under the conditions it is expected to run.

MPC clipping can be easily audible. I came across 1.15r -quality5 --xlevel strong clipping on the Transwave - Helium CD, that is a very well recorded CD. It sounds like vinyl clicks. I don't check anymore all the clicks that I hear in my MPCs. It seems they don't always come from the original CDs. I can find back the sample if you want.

you forgot option D):

- He transcoded

@Pio: My experience of harmless clipping was only in the normal course of things (I routinely use album replaygain), and it's all happened to be during fairly transient-rich sounds (such as orchestra climax and cymbal crash) so that's probably why I didn't notice. I assume then, that in codec testing this was considered not to be a codec-problem if it was just clipping, even though it was audible. I guess inherent the ReplayGain support of Musepack is considered to be the fix.

As for playback on SB resampling soundcards, I agree that it ought to be considered a secondary aim not to have artifacts revealed by them, but if it's rare or requires too many bits to fix, I guess it's the sort of thing that could be ignored (e.g. fb2k with SSRC Resampling is one fix for the problem). We'll have to see if it turns out to be real or not before worrying about it.

OK, I've read some of the posts at Syntrillium (well, as much as I could bare).

I've ripped a sample from the intro to the track teflon20 could apparently ABX with musepack. Phil Collins's, You Can't Hurry Love from the album Serious Hits, Live.

Original sample.

Standard Xlevel sample.

I just tried the samples;
WARNING: I don't want to start a blood war here!
WARNING: I didn't properly ABX it!

I heard both samples about 10 times and I think the second one has a little sharper sound on the clapping (the claps just sound sharper), it happens at second 3-4;

Using headphones I can here the sharp claps in my left year. It might me my imagination but if someone wants to properly ABX those, it might help. Where did the guy say he heard a difference?

[I'm sick of those claps, so no more for me]