Golden Ear test - who is brave enough?

[mtillman]

Here is a 29MB wav of a piece of music.

Some of the file is the original wav ripped from CD by CDex in full paranoia mode, other parts are mp3 encoded using LAME at 128kbps, and yet other parts are mp3 LAME alt preset standard.

So, Golden Ears will consider some parts perfect, other parts crap, and yet other parts OK if you really must use mp3 I suppose.

The file was created in CoolEditPro2, and I will eventually publish a screenshot showing where the edits are.

In the meantime, everyone who's ever claimed to be able to spot an mp3 from 100 yards is invited to tell me what's what. Remember, I know your names

.

The piece I've chosen (which in my opinion is exceptionally beautiful) may or may not be a good choice, but it is very simple and pure, and well recorded, so I would expect that it is a good choice. It's piano and voice, and *very* obscure

That last bit is important, because it *is* possible to cheat at this, but only if you have the complete wav file or the CD. However, even if you do cheat, I still challenge you to distingush between the two different mp3 encodings. BWHAHAHA!!

There ought to be a prize for this. Perhaps a video of me eating your hat?

Any French speakers are requested to provide me with a rough translation

Obviously 29MB is quite a large file so those of you on dial-up are at a considerable disadvantage. I have split the file into five parts using WinRAR if that helps anyone. URLs are:

http://martin.tillman.users.btopenwo...down.part1.rar

http://martin.tillman.users.btopenwo...down.part2.rar

http://martin.tillman.users.btopenwo...down.part3.rar

http://martin.tillman.users.btopenwo...down.part4.rar

http://martin.tillman.users.btopenwo...down.part5.rar


I'm not telling how many edits there are (yet), and I don't expect anyone to be able to identify where the edits are. An approximate timing of which bit is which is sufficient. Oh, and this isn't a double blind test, obviously...


I'm expecting a good response to this, people, and I'm expecting honesty too - from everyone who can't spot the difference - after all, there's no point in fooling ourselves, is there?

For the record, I've been an audio professional for 30 years, the upper limit of my hearing is just over 16KHz, and I can't spot any differences in the file at all. But - I have heard some terrible artifacting in downloaded mp3s.

 

[Blighty]

Quote:

Originally posted by mtillman In the meantime, everyone who's ever claimed to be able to spot an mp3 from 100 yards is invited to tell me what's what. Remember, I know your names . I'm not telling how many edits there are (yet), and I don't expect anyone to be able to identify where the edits are. An approximate timing of which bit is which is sufficient. Oh, and this isn't a double blind test, obviously...

You do realize this is an exceedingly difficult test to pass. Human auditory memory is extremely short. The only method anyone can determine the differences with is by utilizing a DBT with identical recordings. Only under direct comparative listening of short identical sections will someone be able to express which recording quality is better or not. Contrary to popular belief, properly encoded(LAME+EAC) 128kbps is quite good in quality although a properly conducted test can separate 128 from the original WAV file with good headphones or loudspeakers. The threshold where properly encoded mp3s and the original WAV blurs is somewhere around 192kbps. Perhaps if you have very exotic equipment, superhuman hearing and the perfect recording, you could do a little better and move the threshold up to 256kbps but I have as yet to see anyone accomplish the feat in a properly conducted test.

 

[mtillman]

Quote:

Originally posted by Blighty You do realize this is an exceedingly difficult test to pass.

Yes, obviously I do. Well, I hope it is...

Have you seen how many people claim to be able to tell the difference easily?

 

[stallion11msu]

Using a pair of Ety's...sounds absolutely seamless to me. My only guess might be that the first 10 seconds or so are LAME, other than that I'm stumped.

Thanks for doing this mtillman! That piece of music is indeed gorgeous! Now I can rest in confidence that the 300 or so albums I have ripped in LAME -aps are just about completely indistinguishable to my ears from the original.

Peace and Love.

 

[JaZZ]

I can't hear any sonic inhomogeneity within this song – no chance to detect the MP3 sequences. But I must say this cast doesn't represent a critical source to test MP3's weakness: high-frequency transients. If you had chosen any percussion or some synthesizer beats for this test, it would be easier to detect the flaws.

For the encoding (with LAME) of my CDs to load them to my MP3 Jukebox I chose the compression rate according to the audibility of artifacts, mostly the said high-frequency transients corruption, and with good, modern recordings I end up with an encoding rate of at least 256 kbps – 128 kbps is clearly out of consideration.

It's a pity you chose such an example. I'm not one of those who claim MP3 is ****. 256 or at the latest 320 kbps are equal to wave to my ears.

JaZZ


BTW: Obviously it's not French.

 

[bifcake]

The first 10 or 15 secs sound quite inferior to the rest of the piece. In other places, I thought the differences were more subtle. I heard increased compression about half way through and subtle variances later on. However, I am listening to this through SB LIve and computer speakers, so this is by no means a comprehensive test. I'll burn this to a CD and play it on my main rig with Senn 580 phones later.

 

[Tim D]

Yah, some clipping in beginning, and an mp3 or editing artifact at the end at 2:38. Although the first 10 secs clip, it feels like its the original, and seems to get mp3'd at 10-30 seconds with some warble in the piano. Starts to sound like source for awhile but seems to get Mp3'd again after 1:35. But I'm probably missing some transitions or getting some wrong entirely. I think another transition at the 1 minute mark going from Mp3 to original.

 

[AndrewB]

A beautiful piece of music. When you're satisfied with your poll, would you post specifics as to artist and album?

Thanks,
Andrew


ps Other than a few seconds of the entry, sounds just beautiful

 

[PodMan]

i'm on a modem so i'm not going to bother. but I think 128kbps LAME would be what people will be able to pick out easily (as LAME itself is not that optimized for such low bitrates). but can people hear the difference between the standard preset and the original? hopefully, we'll be able to find out.

 

[Blighty]

I burned it onto CD and find that there are definitely artifacts. Whether they were in the original recording, a result of poor extraction, or introduced at the editing phase is inconclusive. They appear multiple times during the first 9 seconds and again at 1:00 and 1:35. Other than that, I have not had much luck determining which passages are mp3 and which are the from original although I have not been listening too long.

 

[Zin_Ramu]

Did not do this test, but have done extensive tests on my rig at home. I first ripped 12 of my favorite songs, then encoded them at -APS. Then burned one compilation disk with the .wav and one with the .mp3. Then I randombly chose disks put them in my sony NS755v and listented through my Corda Ha-1 > Senn HD600. I was unable to determine a difference in a statistically significant manner on any of these 12 tracks.

While some "refrence" samples may trip up LAME, for actual music, I can not tell a difference.

Zin

 

[mekanoplastik]

1:18

 

[dd3mon]

there are definitely artifacts, almost throughout the whole piece.. most evident towards the beginning. If this is evidence of a difference in encoding, poor rip job (use EAC - best there is) or just present in the source I cannot say. I've used EAC+Lame preset standard extensively and I've never heard these type of artifacts introduced from LAME using preset stardard approarch.

-dd3mon

 

[lan]

about 1:10 -> 1:23 is the WAV. the rest is MP3, that's my final answer.

do I win a million dollars!?

 

[averydonovan]

I voted that I couldn't hear the difference since I cannot say for sure if I am really hearing a difference. I have only carefully gone through the first 2 minutes, and here's my guesses (time to make a fool outa myself!):

128Kbps MP3:
0:00 - 0:12
1:25 - 1:46

MP3 (not sure which bitrate):
0:50 - 0:57
1:10 - 1:24

All other sections, up to 1:58 not listed above I am assuming to be the original. As I said above I am not sure if I am really hearing any differences, and I know I don't have golden ears. Also, this type of music is very different from what I normally listen to.

Overall I think the whole thing sounds very good quality-wise. LAME is an excellent encoder, even at lower bitrates. Definately an interesting test