FLAC vs. 320 Mp3
Nov 26, 2012 at 4:52 PM Post #241 of 1,406
Quote:
Show me the science.

 
Not sure what you're actually asking for here.
 
The poster I was answering claimed a difference was audible.  Sound-science is the one area of the forum where it's perfectly OK (it's even preferable) to ask for proof when someone says they've heard a difference.  So I suggested he do an actual abx, and post the log.  Eg show actual data proving that he can tell the difference when testing blind, level matched, and that the results statistically prove his supposition that the difference is audible to him.
 
Does this answer your query - or was that just an off-the-cuff remark?
 
Nov 26, 2012 at 5:12 PM Post #242 of 1,406
Quote:
 
Not sure what you're actually asking for here.
 
The poster I was answering claimed a difference was audible.  Sound-science is the one area of the forum where it's perfectly OK (it's even preferable) to ask for proof when someone says they've heard a difference.  So I suggested he do an actual abx, and post the log.  Eg show actual data proving that he can tell the difference when testing blind, level matched, and that the results statistically prove his supposition that the difference is audible to him.
 
Does this answer your query - or was that just an off-the-cuff remark?

xnor pretty well put me in my place and provided the science.  Interesting stuff, a combination of probability and psychology.
It got me digging a bit.
 
Since I only have a h/s education, I'm having some difficulty with the equations, but I'm handling some of the theories that are fleshed out.
I posted a link http://hal.inria.fr/docs/00/54/47/74/PDF/2007_IEEE_TASLP_OzerovEtAl_SingleChannelSourceSeparation.pdf
What I gathered from some reading is that music is so complex that the source should be mono and the instruments / vocals should be isolated.   I could be wrong.
 
Again limited by my h/s education,  the curves below appear to be diverging as file size increases.  I may be totally wrong about this...
 

 
Nov 26, 2012 at 5:26 PM Post #243 of 1,406
From:
IEEE TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, VOL. 15, NO. 5, JULY 2007
Adaptation of Bayesian Models for "Single-Channel"
Source Separation and its Application to Voice/Music
Separation in Popular Songs 
 
An efficient model must be able to yield a rather accurate description
of a given source or class of sources, in terms of a
collection of spectral shapes corresponding to the various behaviors
that can be observed in the source realizations. This requires
GMMs with a large number of Gaussian functions, which
raises a number of problems:
• trainability  issues linked to the difficulty in gathering and
handling a representative set of examples for the sources
or classes of sources involved in the mix;
• selectivity  issues arising from the fact that the particular
sources in the mix may only span a small range of observations
within the overall possibilities covered by the general
models;
• sensor and channel variability  which may affect to a large
extent the acoustic observations in the mix and cause a
more or less important mismatch with the training conditions;
• computational complexity  which can become intractable
with large source models, as the separation process requires
factorial models [5], [6].
A typical situation which illustrates these difficulties arises
for the separation of voice from music in popular songs. For
such a task, it turns out to be particularly unrealistic to accurately
model the entire population of music sounds with a
tractable and efficient GMM. The problem is all the more acute
as the actual realizations of music sounds within a given song
cover much less acoustic diversity than the general population
of music sounds.
 
Nov 26, 2012 at 5:36 PM Post #244 of 1,406
Even if I can't tell the difference (and in some high quality tracks, I think I can occasionally, but not enough to warrant it), I still rip in FLAC new CD's for piece of mind really.
I then have 320kbps MP3 back ups for when I want to quickly rip tracks to Spotify for transporting over to the phone for listening on the way to work.
 
Lossy files lose the extra information it deems un-neccessary on encoding the file. It doesn't lose any more data. It's not like an LP, or a tape, that can degrade over time.
 
FLAC works cleverly in the fact its a compressed file, that retains all of the encoded information. It just simply uncompresses and decodes the file in real time, that happens to be the quality of the CD originally (if correctly ripped, so no pops or glitches).
 
Exact Audio Copy, which I use has protection mechanisms to check the rip with checksums, to check that the file has remained intact.
 
But yeah, to conclude, if I buy a CD, I will rip it in FLAC for piece of mind, if I can tell the difference or not. I would rather get what I would have got having played using the CD, but I don't tend to keep the CD's in the drive due to noise. I would rather have the file on the hard drive for playing the album, even if it means taking 10 minutes to rip the thing.
 
I won't for example buy MP3's online. If I'm to buy music online, it either has to be in FLAC form, or CD form.
 
Nov 26, 2012 at 7:35 PM Post #245 of 1,406
my sister decided to listen to the Homework album on my iPod Classic 4th gen. along with my Grado Sr80i's
its been CD ripped to iTunesHQ AAC format....
 
 
but compared to her lossy Youtube mp3 audio converts. she says she really hears a difference, and regrets not buying her music in a high quality format.
 
Nov 26, 2012 at 7:51 PM Post #246 of 1,406
I used a high resolution 24-bit 96 kHz ALAC file, "The Coachman's Dance" by Shostakovich, that came as a bonus with my DAC to showcase its performance.  The track is 10:01 long, and the file is 189MB with a bitrate of 2653 kbps.  I converted this file to a 320 kbps CBR mp3 with a file size of 22.9MB following the instructions provided earlier.
 
PC > optical out > NuForce HD Icon as DAC-only with 15V linear power supply > Asgard Amp > Denon D5000 with J$ lambskin ear pads
 
Immediately I noticed that the mp3 file was veiled and the ALAC file had much better detail with superior instrument separation. I could almost sense the humidity level in the air at the auditorium with the ALAC file.  Ok, that is complete BS.
 
I did the Foobar ABX twice.  First up was 6/15, but I attempted to get cute with the file name and cancelled the save to look something up, and apparently lost the log file forever. (Grrrrr!!!!)
 
So I waited an hour, relaxed a bit, and tried the test again.  By this time I was really familiar with the song, and I thought I had a great spot that would showcase the subtle differences between the 2 files.  Nope, just guessing again.
 
 
foo_abx 1.3.4 report
foobar2000 v1.1.13
2012/11/26 18:57:46
 
File A: C:\Music\High Resolution Tracks\HRT_96k_itunes_sample.m4a
File B: C:\Music\High Resolution Tracks\The Coachman's Dance.mp3
 
18:57:46 : Test started.
19:02:01 : 01/01  50.0%
19:07:27 : 01/02  75.0%
19:08:02 : 01/03  87.5%
19:08:36 : 02/04  68.8%
19:09:21 : 02/05  81.3%
19:10:11 : 03/06  65.6%
19:11:25 : 03/07  77.3%
19:12:03 : 03/08  85.5%
19:12:46 : 03/09  91.0%
19:13:30 : 03/10  94.5%
19:14:16 : 04/11  88.7%
19:14:43 : 04/12  92.7%
19:15:31 : 05/13  86.7%
19:16:09 : 05/14  91.0%
19:16:56 : 05/15  94.1%
19:17:01 : Test finished.
 
I can understand, for peace of mind, why many people would want to keep a lossless file available, as nobody has the time or resources to verify that all of their music cannot be identified with an ABX test.  As for me, I'm convinced I am good with lossy mp3 files.  I use lame V 0 320 kbps, which is perfect for my needs, and Google Music uploads these files without transcoding them.  So, my mp3 files are available on Google Music, and I have the CD available too.  I'm not going to bother with lossless archives.
 
I'm also convinced that I can't tell the difference between a CD and a song from either MOG or Spotify when using their highest bitrate streams. 
 
Guess I'm done.  Now, back to enjoying my music.
 
Nov 26, 2012 at 8:13 PM Post #247 of 1,406
Quote:
I used a high resolution 24-bit 96 kHz ALAC file, "The Coachman's Dance" by Shostakovich, that came as a bonus with my DAC to showcase its performance.  The track is 10:01 long, and the file is 189MB with a bitrate of 2653 kbps.  I converted this file to a 320 kbps CBR mp3 with a file size of 22.9MB following the instructions provided earlier.
 
PC > optical out > NuForce HD Icon as DAC-only with 15V linear power supply > Asgard Amp > Denon D5000 with J$ lambskin ear pads
 
Immediately I noticed that the mp3 file was veiled and the ALAC file had much better detail with superior instrument separation. I could almost sense the humidity level in the air at the auditorium with the ALAC file.  Ok, that is complete BS.
 
I did the Foobar ABX twice.  First up was 6/15, but I attempted to get cute with the file name and cancelled the save to look something up, and apparently lost the log file forever. (Grrrrr!!!!)
 
So I waited an hour, relaxed a bit, and tried the test again.  By this time I was really familiar with the song, and I thought I had a great spot that would showcase the subtle differences between the 2 files.  Nope, just guessing again.
 
 
foo_abx 1.3.4 report
foobar2000 v1.1.13
2012/11/26 18:57:46
 
File A: C:\Music\High Resolution Tracks\HRT_96k_itunes_sample.m4a
File B: C:\Music\High Resolution Tracks\The Coachman's Dance.mp3
 
18:57:46 : Test started.
19:02:01 : 01/01  50.0%
19:07:27 : 01/02  75.0%
19:08:02 : 01/03  87.5%
19:08:36 : 02/04  68.8%
19:09:21 : 02/05  81.3%
19:10:11 : 03/06  65.6%
19:11:25 : 03/07  77.3%
19:12:03 : 03/08  85.5%
19:12:46 : 03/09  91.0%
19:13:30 : 03/10  94.5%
19:14:16 : 04/11  88.7%
19:14:43 : 04/12  92.7%
19:15:31 : 05/13  86.7%
19:16:09 : 05/14  91.0%
19:16:56 : 05/15  94.1%
19:17:01 : Test finished.
 
I can understand, for peace of mind, why many people would want to keep a lossless file available, as nobody has the time or resources to verify that all of their music cannot be identified with an ABX test.  As for me, I'm convinced I am good with lossy mp3 files.  I use lame V 0 320 kbps, which is perfect for my needs, and Google Music uploads these files without transcoding them.  So, my mp3 files are available on Google Music, and I have the CD available too.  I'm not going to bother with lossless archives.
 
I'm also convinced that I can't tell the difference between a CD and a song from either MOG or Spotify when using their highest bitrate streams. 
 
Guess I'm done.  Now, back to enjoying my music.

WoW !
 
Nov 26, 2012 at 8:25 PM Post #248 of 1,406
I don`t know, but I seriously don`t believe there is enough of a difference between the two for use on a DAP.  That is, most headphones or even IEMs that people use with DAPs don`t isolate noise well enough to make the FLAC files sound THAT much better than the 320Kbps MP3 files.  For everything else, FLAC makes perfect sense.
 
Nov 26, 2012 at 9:34 PM Post #250 of 1,406
I'm used to Rubinstein, Gilels, Horowitz and Richter.
 
Nov 26, 2012 at 9:36 PM Post #251 of 1,406
ITunes makes lousy MP3 files. Folks should test against AAC 320 or at the very least 320 LAME.
 
Nov 27, 2012 at 12:03 AM Post #252 of 1,406
Quote:
xnor pretty well put me in my place and provided the science.  Interesting stuff, a combination of probability and psychology.
It got me digging a bit.
 
Since I only have a h/s education, I'm having some difficulty with the equations, but I'm handling some of the theories that are fleshed out.
I posted a link http://hal.inria.fr/docs/00/54/47/74/PDF/2007_IEEE_TASLP_OzerovEtAl_SingleChannelSourceSeparation.pdf
What I gathered from some reading is that music is so complex that the source should be mono and the instruments / vocals should be isolated.   I could be wrong.

 
I only really skimmed the abstract, part of the introduction, a few graphs, and parts of the conclusion—as any researcher knows, that's the "proper" way to "read" such documents
wink_face.gif
—but are you sure you linked the right thing?
 
That's a paper describing techniques to deal with this general task: being given one signal (a song recording) and trying to separate one part of it (say the vocals) from the rest.  Pretty much completely unrelated.
 
 
 
Just so we're clear (and at the expense of being needlessly pedantic and insulting everybody's intelligence), here's a description of the actual problem we're looking at.
 
Procedure and reasoning is spoilered below:
 
 
Motivation:
  • Mr. Z says he hear certain differences P and Q between two audio files.
  • Other people wonder if it's true.  Can Mr. Z really hear it?
  • Actually, never mind claims P and Q.  Can Mr. Z hear any difference at all?  If he can't, then the claims about hearing differences P and Q probably aren't true.
  • Let's run an experiment to find out.
 
The experiment, can be done by one person alone with the aid of software like foobar's ABX comparator, or via other means:
  • For the experiment, we will run multiple (e.g. 15) trials.
  • For each trial, Mr. Z listens to the two audio files A and B and then a third option X.  Exact procedure here can vary a bit.
  • X is randomly chosen (unknown to Mr. Z) to be A or B in each trial.  That is, either A or B gets played twice.
  • Mr. Z is asked to identify X as either being a repeat of A or a repeat of B.
  • Count whether or not Mr. Z is correct or not.
  • Repeat until all trials are done.
 
The analysis (simplified, qualitative; hit the books for the statistics, which are not difficult for this simple test):
  • What we wish to test is whether or not Mr. Z is effectively guessing at random (can't tell them apart) or not.  Formally, the hypothesis we tested is that Mr. Z can't tell a difference.
  • If guessing at random, Mr. Z would have a half chance of getting each trial right.
  • If Mr. Z can kind of tell some difference, there would be greater than a half chance of getting each trial right.
  • The higher the percentage of successes in the trials and the higher the number of trials, the more certain we are that Mr. Z can tell a difference.  e.g. If you tossed a coin 100 times and it came up tails 56 times, you wouldn't think anything strange, right?  Same if you tossed once and got tails.  But what if you got tails 93 out of 100 times?  You'd think the coin was rigged—most likely not a fair coin with half probability of tails and half for heads—right?
  • From the results and the statistics, we can crunch a few numbers and get a figure for how likely it was that Mr. Z was guessing (can't tell a difference) or not (can tell a difference).
 
Now what?
  • If Mr. Z could tell a difference in our experiment, then that still doesn't say whether or not he was hearing differences P and Q or something else.  We have more experiments to do.
  • If Mr. Z couldn't tell a difference in the experiment, we can probably say that he wasn't hearing differences P and Q, or whatever, during the experiment.  However, is it possible that Mr. Z could hear differences between the audio files in general but not during the experiment?  Depends on how you run the experiment.  If you want conclusions drawn from the experiment to apply to normal listening circumstances, you probably should run the experiment to mimic normal listening conditions, for example.
 
Nov 28, 2012 at 3:14 AM Post #253 of 1,406
Need some advice:
Simple.... When I rip a cd into iTunes what is the preferred method/setting? Presently I'm choosing AIFF and checked the error correction box.

Anything for iTunes downloads? Again once I get them I right click and choose make AIFF.

I have a Mac so no foobar2000

Thanks
 
Nov 28, 2012 at 3:53 AM Post #254 of 1,406
AIFF is a complete waste. It gobbles up your hard drive and you get nothing for it. Either use the Apple Lossless Encoder or AAC 256 VBR. You won't have any problems with either of those settings. They both sound the same. AAC is better for use with ipods and ipads.

There's no reason to convert iTunes downloads. They're already AAC 256 VBR. Leave em as they are.

It might help you to google basic info so you know how iTunes works.
 

Users who are viewing this thread

Back
Top