[5twnr]

Here's what I ended up using
- ABXTester to test 320 vs v0 (cannot compare flac)
- XLD for FLAC conversion

For the Brandy album, when compared to unknown 320 source, I could tell the difference with 80% accuracy (with 5 trials). I don't have the FLAC for this album.
For the Mariah album, I converted my own 320s and v0s. My accuracy was between 40-60%, so if we average that that's 50%.

I guess with this experiment it is the source that's causing the difference. I haven't found software that would let me set up a test against FLAC though.

Still I'm not sure if I didn't just get listening fatigue; some of the differences are also around soundstage and spacing, and when I'm listening to small segments at a time that tends to get lost. Would need to spend a lot more time to fully test this.

-edit-
Ok now I'm getting 80% on the Mariah track when I listen to half the song for each instead of just snippets.

 

[Brooko]

Need minimum 10 to 15 tests for it to be even close to statistically relevant

 

[gregorio]

[1] For the Brandy album, when compared to unknown 320 source, I could tell the difference with 80% accuracy (with 5 trials).
[2] I guess with this experiment it is the source that's causing the difference.
[3] I haven't found software that would let me set up a test against FLAC though.
[4] Still I'm not sure if I didn't just get listening fatigue; some of the differences are also around soundstage and spacing, and when I'm listening to small segments at a time that tends to get lost. Would need to spend a lot more time to fully test this.
[5] -edit- Ok now I'm getting 80% on the Mariah track when I listen to half the song for each instead of just snippets.

1. As @Brooko mentioned, 5 trials is not enough. If you flipped a coin 5 times, you would not get a 50% heads and tails result. Even with 10 coin flips there's only a moderate probability of getting 5 of each. However, it's not at all surprising to achieve a truly significant result if you don't know the source; the 320 could be a different master or the same master just transcoded from another lossy codec.

2. That's by far the most likely explanation.

3. Just convert the FLAC to WAV (or AIFF) and test with that, you will get a bit perfect copy of the original lossless WAV.

4. It's certainly not impossible that you actually can tell a difference. I could certainly tell the difference between a 320 MP3 and the original in a DBT with 100% accuracy BUT, that was ONLY fast switching carefully selected small segments, with certain pieces of music AND that was quite a number of years ago. This last point is very important because: MP3 is a perceptual lossy codec, meaning that data is removed on the basis of it not being perceivable but how do we know what is not perceivable? Well, we can work out frequencies that will be masked (as explained by castleofargh) and remove them but additionally, thousands of DBT/ABX trials have been completed. Over the years the lossy algorithms have been tweaked numerous times, sparking new rounds of thousands DBT trials, then tweaked again if subjects could identify a difference. Using the more recent versions of MP3 320, I can no longer identify an audible difference. So, if you have good listening skills, know what to look for and fast switch small segments which exhibit what you're looking for, then you can tell the difference between newer and older versions of MP3 320 encoding. It should be mentioned that by about a decade or so ago, virtually all the formal listening testing of the various lossy codecs ceased at the highest bit rates (320 CBR and 256 VBR) because no one could reliably identify a difference any more. Formal listening testing still continues today but at lower bit per second rates, typically 128kbps or lower. So, it's not inconceivable that you can tell a difference between 320 MP3, VBR 256 and lossless (FLAC), if you're using old software (or in theory, newer software which for some reason employs some old lossy algorithms).

5. That flies in the face of everything science has discovered about aural (echoic) memory, everything countless tens of thousands of professional music engineers have witnessed over many decades and everything we've discovered from the millions of ABX/DBT trials done over the last 6+ decades. In other words, what you're suggesting is pretty much inconceivable and the explanation must lie in some fault/failure of your testing procedure, for example not enough trials, not accurately level matched, etc.

G

 

[castleofargh]

Here's what I ended up using
- ABXTester to test 320 vs v0 (cannot compare flac)
- XLD for FLAC conversion

For the Brandy album, when compared to unknown 320 source, I could tell the difference with 80% accuracy (with 5 trials). I don't have the FLAC for this album.
For the Mariah album, I converted my own 320s and v0s. My accuracy was between 40-60%, so if we average that that's 50%.

I guess with this experiment it is the source that's causing the difference. I haven't found software that would let me set up a test against FLAC though.

Still I'm not sure if I didn't just get listening fatigue; some of the differences are also around soundstage and spacing, and when I'm listening to small segments at a time that tends to get lost. Would need to spend a lot more time to fully test this.

-edit-
Ok now I'm getting 80% on the Mariah track when I listen to half the song for each instead of just snippets.

would you send me the files of the track where you converted both versions yourself in PM? just to check if there is an obvious culprit.

and @Brooko is right about stats, you need a relevant number. 15 runs is a fine minimum I'd say, although you are free to reach 15 or 20 over several days at you own pace. what matters is to stick to what you had decided beforehand, and not to start rejecting part of the trial when you don't get a good run or other such funky moves that would completely change the odds. otherwise, you do as you feel works best, after you're trying to pass.

 

[71 dB]

Worrying about the last drops of finesse in miniscule detail is pretty pointless and against the idea of music enjoyment. In my opinion only things you can somewhat easily* hear really matter and differences that take effort to hear are not significant. When allow the sonic quality to be a bit imperfect and you know your system is better than that with some safety margin you can stop worrying about microscopic things and concentrate on the music itself. Soundstage 3 % flatter? Did that stop you enjoying the chord progression, melodies, rhythms and singing?

* Being 80 % of the time correct doesn't mean easily heard differences. It means you need to think why you were incorrect 20 % of the time? Lets say easily easily heard differencies causes incorrect (because we are humans) answers 1 % of the time at MOST. That requires a success rate of 99 % or better!

 

[castleofargh]

5. That flies in the face of everything science has discovered about aural (echoic) memory, everything countless tens of thousands of professional music engineers have witnessed over many decades and everything we've discovered from the millions of ABX/DBT trials done over the last 6+ decades. In other words, what you're suggesting is pretty much inconceivable and the explanation must lie in some fault/failure of your testing procedure, for example not enough trials, not accurately level matched, etc.

 

[castleofargh]

@5twnr sent me a CBR and VBR version of a song, and I saw nothing obviously wrong with those. both say 44.1khz playback, no massive clipping(max true peak reaches +0.6dB on both), no volume difference, latest LAME encoder(is it like saying ATM machine?), same duration(to the last sample). DiffMaker went below -60dB so at first glance the files seem legit to me and properly converted.

I tried an ABX and after 3mn of practice I thought there was some small difference in a sort of panning yoyo for low freq sounds, but when I ran the abx on that portion of music, I failed to identify anything. I didn't spend a lot of time on this and am not familiar with that song, but right now I'm not really confident that I could pass an abx on my system if I had more time.

 

[Lulu800]

Has anyone here did stuff with Experimental and industrial?. I've noticed quite few artefacts with AAC/vorbis from 128 to 320kbps, Only 3+ with Lame V0/320k.

 

[bigshot]

The type of music shouldn't affect how well a compression codec works. There are certain sounds, like audience applause and massed strings, that don't encode as easily as other kinds of sound, but all kinds of music can contain those sorts of sounds.

 

[Lulu800]

The type of music shouldn't affect how well a compression codec works. There are certain sounds, like audience applause and massed strings, that don't encode as easily as other kinds of sound, but all kinds of music can contain those sorts of sounds.

True seem's like big 3 culprits are synth's based noises, distorted ambient noises, complex/chaotic music with no pauses. Those show up even less experimental stuff, like some metal/rock.

 

[bigshot]

Are you sure what you are interpreting as artifacts aren't part of the synth voice? Generally artifacts sound like synthetic kinds of sounds, so they stand out most over acoustic instruments that we know what they are supposed to sound like in real life.

 

[Lulu800]

I've did 15 tests with them, they don't show up on any of my lossless content. Like hearing puffy wind noise with vorbis when a loud synth noise is there. Skin crimes - Heaven gate intro had that issue its really apparent with my ER4SR's on 160kbps ogg, it dosen't go away untill the bitrate is 420kbps.

 

[bigshot]

Could you please dropbox me a lossless copy of that track to play with? PM me and I'll give you my email addie.

 

[Lulu800]

Could you please dropbox me a lossless copy of that track to play with? PM me and I'll give you my email addie.

Sen't a PM.

 

[TheTrace]

How come no one talks about opus?

I convert everything from FLAC to opus 192 kbps. Never have a problem. Used to have AAC 256 before I changed to Android