Is there any academic or somewhat scientific study showing that 320 kbps aac is audibly different from lossless?

Oct 11, 2015 at 2:19 PM Thread Starter Post #1 of 16

AutumnCrown

100+ Head-Fier
Joined
Sep 24, 2015
Posts
262
Likes
72
Basically what the title says. I ask because a few studies have shown marginal differences between lossless and 320 kbps mp3 and lossless, for example this one where only engineers were able to hear the difference, and only slightly better than guessing then:
 
http://www.academia.edu/441306/Subjective_Evaluation_of_MP3_Compression_for_Different_Musical_Genres
 
However, considering how much better aac is compared to mp3 at lower bitrates, it seems likely that very small differences between lossless and lossy mp3 would disappear on 320 kbps aac.
 
Can anyone provide any evidence that 320 kbps aac is different from lossless?
 
Thanks
 
Oct 11, 2015 at 5:03 PM Post #2 of 16
Oct 11, 2015 at 5:56 PM Post #3 of 16

So I guess the answer is "no, because scientists aren't doing that anymore"?
 
Unfortunately, that forum, like most kbps blind testing resources, refers to the mp3 format more than aac.
 
I am also interested in non-academic tests if they seemed well done. But to be clear, by test I do mean something with blinds and controls, etc.
 
Oct 11, 2015 at 8:22 PM Post #4 of 16
Here's a list of some codec tests:
https://en.wikipedia.org/wiki/Codec_listening_test
 
Not sure if any one in particular answers your question. This quote from the above-linked hydrogen thread seems to sum it all up pretty well:
 
Code:
 [color=rgb(34, 34, 34)]"The following is a generalization:[/color] [color=rgb(34, 34, 34)]Samples that can be ABXed at 320 kbps by people trained to hear artifacts are fairly rare.[/color] [color=rgb(34, 34, 34)]Samples that can be ABXed at 320 kbps by people who are not trained to hear artifacts are (virtually) nonexistent.[/color] [color=rgb(34, 34, 34)]This is what I have concluded from years of reading what is being said here at HA, but as far as I know nobody has gone back to compile this information in any kind of quantitative way."[/color]
 
If someone with lots of experience listening to claves heard a small difference in the sound of claves between 256AAC and WAV in a controlled, quiet listening environment, one side would say "well just shows people really don't hear much difference", while the other side would say "SEE, LOSSY COMPRESSION RUINS THE MUSIC."
 
Oct 11, 2015 at 8:38 PM Post #5 of 16
Thanks for your response. I've seen that list before. Unfortunately, all of them are lower than 320 kbps.
 
Just to reiterate, I suspect that there may be a significant difference between 320 mp3 and aac - so the quote doesn't really get to an important facet of the issue, imo.
 
I also saw that quote earlier, but I was hoping that there would be a study that could lay it to rest once and for all, and demonstrate that the difference between all 320 kbps aac and flac audio is above every human's perception. Maybe it doesn't exist yet, or never will.
 
Oct 11, 2015 at 8:59 PM Post #6 of 16
  Thanks for your response. I've seen that list before. Unfortunately, all of them are lower than 320 kbps.
 
Just to reiterate, I suspect that there may be a significant difference between 320 mp3 and aac - so the quote doesn't really get to an important facet of the issue, imo.
 
I also saw that quote earlier, but I was hoping that there would be a study that could lay it to rest once and for all, and demonstrate that the difference between all 320 kbps aac and flac audio is above every human's perception. Maybe it doesn't exist yet, or never will.

 
The spirit of the quote would be that the set of "killer" samples for 320mp3 is larger than the set for 256AAC, and I imagine few would have much issue with such a claim. Whether AAC can always keep up with FLAC for everyone is another question. You only need one sample where one person can reliably spot a difference, but Apple sure isn't going to make a big deal about it if they find such a sample+person pair.
 
Oct 11, 2015 at 11:59 PM Post #7 of 16

I would be interested in hearing what it would sound like if a lossless track was inverted and overlaid with a 320 kbps aac. That might provide some insight into the magnitude of the lost data in terms of human hearing...
 
anyone with the necessary skills want to do that for me? 
bigsmile_face.gif
 
 
Oct 12, 2015 at 12:20 AM Post #8 of 16
null/difference testing is not valid for perceptual codec evaluation - the whole point is to throw out information about stuff that can't be heard due to the sounds that can be
 
repeat  - the stuff tossed is inaudible while the rest of the music is playing - the fact that it can be heard on its own doesn't tell you what you expect
 
Oct 12, 2015 at 1:42 AM Post #9 of 16
  null/difference testing is not valid for perceptual codec evaluation - the whole point is to throw out information about stuff that can't be heard due to the sounds that can be
 
repeat  - the stuff tossed is inaudible while the rest of the music is playing - the fact that it can be heard on its own doesn't tell you what you expect


Got it
 
Oct 12, 2015 at 4:33 AM Post #10 of 16
I've always seen ACC vs MP3 the same way I was seeing OGG vs MP3 at the beginning of it all. the purpose of such codecs is to save as much space as possible without changing the audible sound too much, so one is given as superior when it does that, not when it's the most transparent.
AAC is clearly superior to MP3 at lower bitrates in that sens. we save a lot of space and still end up with subjectively better sound. but at the higher bitrates, as the actual audible differences melt like judge Doom at the end of "who framed roger rabbit", it really becomes a question of space more than anything else.
 
Oct 13, 2015 at 1:59 PM Post #11 of 16
I've searched for this myself and could find no reliable evidence to suggest that anyone could hear a difference.  When I convert the files for testing, or use files that I am confident have been properly encoded, I am unable to hear a difference.  Now my statements will be interpreted by some to mean that every amp and DAC sounds the same and that lossy and lossless files are identical, and I will be asked to provide proof that this is true. 
 
The only thing I found is that the MP3 codec can have problems with transients due to the limited blocksize of 192, which may be responsible for pre-echo or ringing sometimes heard by a select few individuals.  This could be the area where MP3 has difficulties with cymbals and similar high frequency sounds.  I would never call this a night and day difference, but audiophiles that would probably never even be able to hear this affected frequency range at all will join in the fun and make claims that their expensive gear reveals this to be similar to cats having sex in the alley outside their open window.  In any event, the AAC codec is more accurate in this regard and uses a smaller blocksize that would seem to alleviate this potential issue.
 
http://www.diffen.com/difference/AAC_vs_MP3 
 
Something else to consider is that there seems to be a difference in how iTunes creates AAC files when compared to Fraunhofer FDK AAC software, at least with regards to the default settings when creating 320 kbps files.  It was discovered that the Tidal music online test had measurable differences between their FLAC and AAC 320 files, and these could be heard in an ABX test, even by me.  I had posted my ABX 2.0 results in one of the Tidal threads here.  The same differences found in the Tidal AAC files could be recreated using dBPoweramp and the Fraunhofer FDK ACC codec download.  This issue was not present with any AAC files converted with iTunes tools.
 
Few people seem to want to go through any rigorous steps when testing, and because there are so many factors that could be responsible for differences being heard, a lot of people just assume they can hear differences and move on.  For me, I started out saying to myself that there is absolutely no way these files sound the same.  I thought that I could easily hear differences and that the lossless files sounded much better.  It was so convincing that I had to research and verify that something in my test setup was not forcing the 2 files to sound the same when I did an ABX.  For months while listening to lossy files, anytime I heard some anomaly of any kind that seemed wrong or out of place, every ABX test had a null result.  What I was hearing was always present on the CD and I just never noticed it before.
 
I progressed from file format to hardware after my epiphany.  There simply does not seem to be a lot there to make the music sound different, and since our ears adjust to what we are hearing, there just didn't seem to be any point in spending any time tweaking for improvement after a certain level of quality was achieved.  People would be better off worried about the barometer reading or humidity levels, or how much liquid was sitting in their cup on the desk, as this might actually be making a bigger difference than the other stuff they think should be.  
 
Oct 13, 2015 at 11:01 PM Post #12 of 16
   
Something else to consider is that there seems to be a difference in how iTunes creates AAC files when compared to Fraunhofer FDK AAC software, at least with regards to the default settings when creating 320 kbps files.  It was discovered that the Tidal music online test had measurable differences between their FLAC and AAC 320 files, and these could be heard in an ABX test, even by me.  I had posted my ABX 2.0 results in one of the Tidal threads here.  The same differences found in the Tidal AAC files could be recreated using dBPoweramp and the Fraunhofer FDK ACC codec download.  This issue was not present with any AAC files converted with iTunes tools.
 

 
I know that the "tidal test" was eq'd and to make the difference between lossless and and acc audible. Is what you are referring something with all Tidal aac, or just the 5 song test?
 
Oct 14, 2015 at 10:37 AM Post #13 of 16
   
I know that the "tidal test" was eq'd and to make the difference between lossless and and acc audible. Is what you are referring something with all Tidal aac, or just the 5 song test?

 
It is only the Tidal test I am referring to, as I have no data to verify the entire Tidal catalog, but I'm not convinced the differences present were intentionally created.
 
This is where I started my journey that led me to believe that the Fraunhofer FDK ACC encoding software was responsible for the "EQ" presence in the Tidal test files.
 
https://www.reddit.com/r/headphones/comments/2n4zm5/my_apologies_to_tidal_with_a_caveat/ 
 
I've been meaning to post ABX results about this issue.  I have the CD of the Eagles Hotel California version that was used in the test.
 
I want to create an AAC 320 file with dBPoweramp and Fraunhofer, another AAC 320 file with iTunes, a Lame version 3.99 MP3 vbr -0 file with dBPoweramp, and a FLAC version using dBPoweramp.
 
I have a copy of Tidal's FLAC and AAC 320 test files, and I can use my own files and ABX various scenarios to see if any differences exist. (obviously only to me)
 
When I tried this before, the Tidal FLAC, my FLAC, my Lame MP3 vbr -0, and my my iTunes created AAC 320 could not be statistically proven by me to sound different from each other.  I have software (of questionable legality) that extracts the Google Music streams, which are Lame version 3.98 320 CBR files, and this was also audibly transparent with these other files.  Only the Tidal AAC 320 sounded different to me, and I could ace any ABX test that I tried up to 15/15.  
 
I tested 10/10 and posted the results (took forever for me to find the post):
 
http://www.head-fi.org/t/770352/how-well-can-you-hear-audio-quality#post_11662420
 
Oct 15, 2015 at 3:29 AM Post #14 of 16
I've never heard of any academic studies being done.  I just took some music I have on CD which I know very well, created a FLAC copy with dBpoweramp.  Then I took foobar2000 and encoded the song to AAC, MP3 and Ogg Vorbis and tried to see if I could tell the difference.  I could not.  FDK-AAC -Q5 is good enough quality for me.  It will create bigger files than if I set the quality setting lower, but I have enough space.  When I start running out, I'll probably drop it lower.  I keep lossless around for backup.
 
Oct 16, 2015 at 1:51 AM Post #15 of 16
   
It is only the Tidal test I am referring to, as I have no data to verify the entire Tidal catalog, but I'm not convinced the differences present were intentionally created.
 
This is where I started my journey that led me to believe that the Fraunhofer FDK ACC encoding software was responsible for the "EQ" presence in the Tidal test files.
 
https://www.reddit.com/r/headphones/comments/2n4zm5/my_apologies_to_tidal_with_a_caveat/ 
 
I've been meaning to post ABX results about this issue.  I have the CD of the Eagles Hotel California version that was used in the test.
 
I want to create an AAC 320 file with dBPoweramp and Fraunhofer, another AAC 320 file with iTunes, a Lame version 3.99 MP3 vbr -0 file with dBPoweramp, and a FLAC version using dBPoweramp.
 
I have a copy of Tidal's FLAC and AAC 320 test files, and I can use my own files and ABX various scenarios to see if any differences exist. (obviously only to me)
 
When I tried this before, the Tidal FLAC, my FLAC, my Lame MP3 vbr -0, and my my iTunes created AAC 320 could not be statistically proven by me to sound different from each other.  I have software (of questionable legality) that extracts the Google Music streams, which are Lame version 3.98 320 CBR files, and this was also audibly transparent with these other files.  Only the Tidal AAC 320 sounded different to me, and I could ace any ABX test that I tried up to 15/15.  
 
I tested 10/10 and posted the results (took forever for me to find the post):
 
http://www.head-fi.org/t/770352/how-well-can-you-hear-audio-quality#post_11662420


Thanks for all the info. I am not sure I understand it totally. How would you describe the audible differences in the tidal test? Is it just a matter of amplitudes of certain frequencies?
 

Users who are viewing this thread

Back
Top