YouTube Audio (and ABX)

[Fran Guidry]

What do you think of YouTube audio? Have you noticed any improvements lately?

I extracted the audio from one of my clips both before I uploaded it and after YouTube compressed it and downloaded it. The upload was a 48 khz PCM and the download was a 152 kbps AAC. The compression ratio is about 10 to 1.

I blinded the files by converting the AAC to the same 48 khz PCM format as the original. Here they are:

88.wav
99.wav

These are large files, you'll probably find them easier to play if you download them to your system. Can you hear a difference?

In past comparisons I've given out the key privately to anyone who asked for it, but this time I'm asking for a little more. I'll only provide the key to folks who convince me that they've ABXed the files and either can't tell a difference or did succeed in 13 out of 16 trials. I'm doing this because I'd really like to know if there's a difference that people can hear, since my ears are old and damaged I'm not willing to accept my own ability to discriminate.

I did a blog post about this whole mess, with more commentary on upload and download formats: http://www.homebrewedmusic.com/2012/...audio-and-abx/

And for PC users I did a blog post and video with more info on the PC ABX tool that works with the foobar2000 audio player: http://www.homebrewedmusic.com/2009/...dio-interface/

Fran

[Howard Emerson]

Hi Fran,
I went back and forth between the two several times, listening at low volume on my computer speakers.

They are close enough that whichever one took less work to do is the better one.

Great left hand, Fran!

HE

[Fran Guidry]

Thanks, Howard. Just trying to do Uncle Sonny justice.

This latest round of upgrades at YouTube seem to have gotten us pretty close to transparency. Not too long ago any audio from YT was burbling and grainy with compression artifacts, but I'm not hearing that in these clips.

Fran

[mesa]

.....sorry wrong thread!

[Rick Shepherd]

Originally Posted by Fran Guidry

Thanks, Howard. Just trying to do Uncle Sonny justice.

This latest round of upgrades at YouTube seem to have gotten us pretty close to transparency. Not too long ago any audio from YT was burbling and grainy with compression artifacts, but I'm not hearing that in these clips.

Fran

I notice and significant difference between the two.

[Scott Whigham]

Love this - and thanks for all this work, Fran. I do hear differences but right now I'm on a desktop computer. I'll fire these up in the studio tomorrow and make a guess!

Very cool - thanks again.

[Doug Young]

I tried this, and I feel like I can hear a difference between the two files, tho it's so close that I'd certainly go to no effort to hear one over the other. However, just to get past the "I know what I hear..." thing, I did the ABX test. The 1st time I scored 11/16 correct 68%, which might seem like I really am hearing something - only 2 tries short of your criteria! But a look at the log tends to indicate I might be guessing :-) I did seem to pick up speed at the end, tho, - the last 4 were all correct - so maybe I was just warming up. So I did it a second time - I should get better with practice and tune in to the differences, right? And I felt like I was improving, I was much more sure I was correctly identifying the right ones. But hmm, 2nd pass, I scored 50%, what do you know, I am guessing :-)


Here's my ABX log:

ABX Test Completed: 2012-03-18 14:55:13 -0700

Number of tests performed: 16
Number of correct answers: 11
Percentage correct: 68%

File 1 = /Users/user/Downloads/99.wav
File 2 = /Users/user/Downloads/88.wav
File placement was static.

n [A] [X] [B] Choice Score
1 [2] [1] [1] B 1/1
2 [2] [1] [1] A 1/2
3 [2] [2] [1] A 2/3
4 [2] [2] [1] B 2/4
5 [2] [1] [1] B 3/5
6 [2] [2] [1] B 3/6
7 [2] [2] [1] A 4/7
8 [2] [2] [1] A 5/8
9 [2] [2] [1] A 6/9
10 [2] [2] [1] B 6/10
11 [2] [1] [1] B 7/11
12 [2] [1] [1] B 8/12
13 [2] [1] [1] A 8/13
14 [2] [1] [1] B 9/14
15 [2] [1] [1] B 10/15
16 [2] [1] [1] B 11/16

--------------------------------------------------------------

ABX Test Completed: 2012-03-18 15:27:40 -0700

Number of tests performed: 16
Number of correct answers: 8
Percentage correct: 50%

File 1 = /Users/user/Downloads/99.wav
File 2 = /Users/user/Downloads/88.wav
File placement was static.

n [A] [X] [B] Choice Score
1 [2] [1] [1] A 0/1
2 [2] [1] [1] B 1/2
3 [2] [2] [1] A 2/3
4 [2] [2] [1] B 2/4
5 [2] [1] [1] A 2/5
6 [2] [1] [1] B 3/6
7 [2] [2] [1] A 4/7
8 [2] [1] [1] A 4/8
9 [2] [2] [1] A 5/9
10 [2] [1] [1] A 5/10
11 [2] [2] [1] B 5/11
12 [2] [1] [1] B 6/12
13 [2] [2] [1] A 7/13
14 [2] [2] [1] B 7/14
15 [2] [2] [1] A 8/15
16 [2] [2] [1] B 8/16

[moon]

15/16 on ABX. The mistake was on my second try but after that I'd focussed in on the difference and got it right every time. There's some burbling in the 99 file, although not a lot. The better of the two files sounded more "solid". I would guess this is some kind of timing thing where all the right frequencies peak at just the right moment rather than being smeared or blurred. There's probably a technical term for this but maybe you know what I mean.

I listened to the files on an RME Multiface & Sony MDR V6, which I'd class as mid-range audio gear. Anything south of that and I doubt if I could tell them apart. The lesser file is probably good enough unless you've got your computer hooked up to a good sound system.

[Howard Emerson]

Originally Posted by Doug Young

I tried this, and I feel like I can hear a difference between the two files, tho it's so close that I'd certainly go to no effort to hear one over the other. However, just to get past the "I know what I hear..." thing, I did the ABX test. The 1st time I scored 11/16 correct 68%, which might seem like I really am hearing something - only 2 tries short of your criteria! But a look at the log tends to indicate I might be guessing :-) I did seem to pick up speed at the end, tho, - the last 4 were all correct - so maybe I was just warming up. So I did it a second time - I should get better with practice and tune in to the differences, right? And I felt like I was improving, I was much more sure I was correctly identifying the right ones. But hmm, 2nd pass, I scored 50%, what do you know, I am guessing :-)


Here's my ABX log:

ABX Test Completed: 2012-03-18 14:55:13 -0700

Number of tests performed: 16
Number of correct answers: 11
Percentage correct: 68%

File 1 = /Users/user/Downloads/99.wav
File 2 = /Users/user/Downloads/88.wav
File placement was static.

n [A] [X] [B] Choice Score
1 [2] [1] [1] B 1/1
2 [2] [1] [1] A 1/2
3 [2] [2] [1] A 2/3
4 [2] [2] [1] B 2/4
5 [2] [1] [1] B 3/5
6 [2] [2] [1] B 3/6
7 [2] [2] [1] A 4/7
8 [2] [2] [1] A 5/8
9 [2] [2] [1] A 6/9
10 [2] [2] [1] B 6/10
11 [2] [1] [1] B 7/11
12 [2] [1] [1] B 8/12
13 [2] [1] [1] A 8/13
14 [2] [1] [1] B 9/14
15 [2] [1] [1] B 10/15
16 [2] [1] [1] B 11/16

--------------------------------------------------------------

ABX Test Completed: 2012-03-18 15:27:40 -0700

Number of tests performed: 16
Number of correct answers: 8
Percentage correct: 50%

File 1 = /Users/user/Downloads/99.wav
File 2 = /Users/user/Downloads/88.wav
File placement was static.

n [A] [X] [B] Choice Score
1 [2] [1] [1] A 0/1
2 [2] [1] [1] B 1/2
3 [2] [2] [1] A 2/3
4 [2] [2] [1] B 2/4
5 [2] [1] [1] A 2/5
6 [2] [1] [1] B 3/6
7 [2] [2] [1] A 4/7
8 [2] [1] [1] A 4/8
9 [2] [2] [1] A 5/9
10 [2] [1] [1] A 5/10
11 [2] [2] [1] B 5/11
12 [2] [1] [1] B 6/12
13 [2] [2] [1] A 7/13
14 [2] [2] [1] B 7/14
15 [2] [2] [1] A 8/15
16 [2] [2] [1] B 8/16

Hi Doug,
Whew...........Glad I stuck to my method:-)

HE

[Doug Young]

Originally Posted by moon

There's some burbling in the 99 file, although not a lot. The better of the two files sounded more "solid".

"gurgley" is the word I used in offline discussion with Fran for what I hear on my own You Tube videos. "burbling" seems right too :-). I'm not finding it in this example, tho, for some reason.

Howard, that's just the output trace of the ABX program I used. It's quite simple to use, just listen and click a button to vote. The funny thing about ABX is that it sometimes works both ways. There have been times when I think I hear a difference, but can't find it when I have to choose blind, and other times when the process lets me zero in on an obvious difference that I would not have noticed with casual listening. Either way, it's always educational.

[Fran Guidry]

Originally Posted by Doug Young

I tried this, and I feel like I can hear a difference between the two files, tho it's so close that I'd certainly go to no effort to hear one over the other. However, just to get past the "I know what I hear..." thing, I did the ABX test. The 1st time I scored 11/16 correct 68%, which might seem like I really am hearing something - only 2 tries short of your criteria! But a look at the log tends to indicate I might be guessing :-) I did seem to pick up speed at the end, tho, - the last 4 were all correct - so maybe I was just warming up. So I did it a second time - I should get better with practice and tune in to the differences, right? And I felt like I was improving, I was much more sure I was correctly identifying the right ones. But hmm, 2nd pass, I scored 50%, what do you know, I am guessing :-)

...

Thank you, Doug. The ins and outs of statistics are funny, and I'm far from an expert and even farther from wanting to be, but it seems that if you feel like you're getting better, just keep going. It will take more trials but once you get a run of correct guesses your stats will go into the range indicating certainty.

I've read a bit and found this table: http://home.provide.net/~djcarlst/abx_bino.htm It indicates that your first run of 11 out of 16 could reasonably be expected to occur 1 time out 10 using pure guesswork. Getting two more right, 13 out of 16, bumps that up to 1 time out 100.

Here's one bump in the road, though. It takes a lot of us failing to provide convincing evidence that there's no difference. It turns out to be a tough task, one that takes a lot of concentration and energy, one that relies on finding the relevant artifact before attention flags and performance falls off.

That doesn't eliminate the other big bump - saying that one hears a difference is completely and utterly different from demonstrating that one hears a difference.

Fran

[Fran Guidry]

Originally Posted by moon

15/16 on ABX. The mistake was on my second try but after that I'd focussed in on the difference and got it right every time. There's some burbling in the 99 file, although not a lot. The better of the two files sounded more "solid". I would guess this is some kind of timing thing where all the right frequencies peak at just the right moment rather than being smeared or blurred. There's probably a technical term for this but maybe you know what I mean.

I listened to the files on an RME Multiface & Sony MDR V6, which I'd class as mid-range audio gear. Anything south of that and I doubt if I could tell them apart. The lesser file is probably good enough unless you've got your computer hooked up to a good sound system.

Thanks very much for participating, moon.

Finding the artifact can be tricky but if it's there and you lock on to it the task gets a lot easier. Can you point us to a section where the artifacts are most easily noticed?

I'll PM you the key.

Fran

[Fran Guidry]

Originally Posted by Doug Young

"gurgley" is the word I used in offline discussion with Fran for what I hear on my own You Tube videos. "burbling" seems right too :-). I'm not finding it in this example, tho, for some reason.

Howard, that's just the output trace of the ABX program I used. It's quite simple to use, just listen and click a button to vote. The funny thing about ABX is that it sometimes works both ways. There have been times when I think I hear a difference, but can't find it when I have to choose blind, and other times when the process lets me zero in on an obvious difference that I would not have noticed with casual listening. Either way, it's always educational.

I put these clips up on the Acoustic Guitar Mag board and a poster there pointed me to the first bass note and a difference in "intensity." Once he pointed it out I've been able to ABX the tracks consistently. And he noted that he didn't hear the difference until he used ABX to put the clips under the attention microscope.

Fran

[Doug Young]

Originally Posted by Fran Guidry

It takes a lot of us failing to provide convincing evidence that there's no difference.

To be clear, you don't mean there's no difference (that's easy to test with a NULL), just that difference is below the threshold of hearing of most or many people, right?. I'm not sure there can be answer to that, because there are different ears and different gear we're listening on, so you and I might never hear the difference, but some young kid with un-rock n' rolled ears and a fantastic studio setup might hear it easily. I suspect that You Tube believes the difference is below the threshold of hearing (or caring) for the bulk of their listeners, and they're probably right.

[Lpendes]

I remember my audio tech teacher talking about how back in the day of myspace, he used to master tracks specifically for Myspace. They would run tests by uploading and then downloading tracks and in the end, the master track that came out of whatever DAW they used, would sound like s**t! But once uploaded to Myspace it would sound great!