Can You Really Hear the Difference Between Lossless, Lossy Audio?

CWmike writes "Lossless audio formats that retain the sound quality of original recordings while also offering some compression for data storage are being championed by musicians like Neil Young and Dave Grohl, who say compressed formats like the MP3s being sold on iTunes rob listeners of the artist's intent. By Young's estimation, CDs can only offer about 15% of the data that was in a master sound track, and when you compress that CD into a lossy MP3 or AAC file format, you lose even more of the depth and quality of a recording. Audiophiles, who have long remained loyal to vinyl albums, are also adopting the lossless formats, some of the most popular of which are FLAC and AIFF, and in some cases can build up terabyte-sized album collections as the formats are still about five times the size of compressed audio files. Even so, digital music sites like HDtracks claim about three hundred thousand people visit each month to purchase hi-def music. And for music purists, some of whom are convinced there's a significant difference in sound quality, listening to lossy file formats in place of lossless is like settling for a Volkswagen instead of a Ferrari."

Comments

Any good studies?

[Experiment 626]

Anyone know of any good double-blind studies comparing people's ability to tell FLAC from 320kbps MP3? Googling just turns up people debating in forums whether you would be able to tell the difference rather than any serious academic research.

Re:Any good studies?

[FrankSchwab]

I don't know if it's a good study, but I did exactly this test. Ten or fifteen years ago.

I took four musical selections (from the latest Rolling Stones album at the time, a solo piano performance, a classical orchestra, a female vocal), and encoded them at 128, 192, and 256 Kbps with the Fraunhofer codec of the day (remember that?). I re-expanded them to 44.1 KHz CD tracks, and put them on a burned audio CD (remember those?). Each selection on the CD had five versions - the first was always the original bit-for-bit copy from the source CD, then followed (in random order) the 128, 192, 256, and the original again.

I made ten copies, and handed them out to the audiophiles in the office to play on their home stereos, and gave them a test sheet - I asked them to identify for each selection which version was 128, 192, 256, or the original. Nobody came close to having a "golden ear" that could reliably tell the 128Kbps versions from the others, much less the higher bitrates. Overall, there was a slight ability to detect the 128 kbps versions - it got selected as the lowest quality one more times than random chance would suggest, but even it was still well below 50% (I don't remember the exact numbers any more).

And this was with ancient MP3 encoders.

Frankly, if you think you've got the golden ear, first of all I pity you - I'm sorry that you have to put up with all the crap you're going to hear. Second of all, I really recommend running the same test - prepare the tracks, have a friend randomly order them (but keep track), and then see if you can identify them. Don't simply say "Of course I can" - Actually do it and prove it.

And, if I can be an old man with a bit of advice for a minute: if you can't tell the difference, don't go out of your way to train yourself to tell the difference. It'll just be an annoyance to you for the rest of your life. Kinda like the person who taught me about the reel-change indicators on film at the movie theatres - I see it, and my whole body tenses up waiting for the change. I wish I had never known about it. I really appreciate the change to digital projection so I don't have to deal with those anymore. /frank

Re:Depends on the source

[fa2k]

Depends on how good the sound engineers are. A lot can be gained by higher resolution and sample rate in the mastering stage, but by using a good low pass filter and dithering (and dithering is not really necessary, http://developers.slashdot.org/story/13/02/27/1547244/xiph-episode-2-digital-show-tell ) basically all audible information is captured in 44.1kHz / 16. Your speakers probably don't go much above 20 kHz anyway, so anything beyond 44.1kHz will only cause distortion (aliasing), see post by MetalliQaZ "Debunked" below.

Re:Depends on the source

[hairyfeet]

You are 100% correct, I have sat in a $100k studio with $5k reference monitors and heard my tracks played back at both 192k and at 44.1k and honestly? Couldn't tell the difference, i really couldn't. And while my midrange hearing may not be the greatest I'm picky as hell when it comes to low end and that is usually the first thing that goes when you compress but standard 44.1k? Couldn't tell the difference which if there was gonna be a difference i would have heard it on that system, it was top notch. I'm sure many here can bring citations showing double blind tests which i have no doubt show its all placebo, because if I can't hear it in a nice studio with the actual live instrument right beside it i doubt seriously anybody is gonna hear a difference with home gear, even high end home gear.

Re:Depends on the bitrate

[jlfose]

It could be dependent on the gear that playback occurs on and the quality of the listener's ears. In watching Stan Lee's new show about "superhumans" it becomes clear that some people have, by training or genetics, better reflexes then the bulk of humanity. On my home gear I can't tell the difference above 160Kbs, but I'm more then willing to believe that some people can, either because they have much better gear to listen to, and/or they have superior hearing.

Re:Depends on the source

[hairyfeet]

Well to be really REALLY fair I have noticed it also matter if the original music was recorded in analog or digital, as I've taken some tracks we've cut in a classic studio with the analog 8-track and its really fricking hard to get those to sound really..."right" for want of a better term as its really hard to describe, when it is compared to digital.

The closest I can get to describing it is this and sorry if you aren't a musician but they'll know of which I speak...you know how you have that great old tube amp for the guitar and it has that nice warm fat feel to it? Notice how the same amp when modeled digitally doesn't doesn't quite have the warmth? Its kinda...artificial sounding? That was the trouble we had, the tapes sounded nice and warm but trying to get that to switch over to digital was fucking hard, frankly it was easier to just cut the tracks again in a digital studio than it was to get the analog tapes to really convert well.

Sorry if I'm not describing it correctly but music is one of those things where my terminology often fails me, its so hard to describe feelings and emotions and music for me is an emotional expression so I end up having to try to describe how I felt as I listened or played and my vocabulary fails me, the analog was a little fuzzy but it was warm and lived in feeling while trying to convert that to digital something was lost in translation, no other way I know how to say it. the same tracks recorded natively in digital sounded great, analog sounded great, but putting the two together was just something we never could get to work.

Re:AIFF?, Flac!, Lossless in General. & Random

[evilviper]

we devised a simple test. He would show me two identical-looking files in iTunes, just showing the titles. One was a high-bitrate AAC and the other a FLAC file. [...] I got 9 of 10 right.

AAC (like MP3) is a frequency-domain codec, and can therefore never provide transparent audio. It has nothing to do with "deeper". but instead is an inability to represent transients... non-tonal components like percussive sounds and other noise.

If you had performed the test with Musepack/MPC or even MPEG-1 Layer II at high bitrates, you would have failed the test.

http://en.wikipedia.org/wiki/MPEG-1#Quality

Re:Depends on the source

[Entropius]

OT, as a choral performer:

Classical music has a stupid wide dynamic range, more than any other genre I know of, and (in particular) soprano sections have a nasty talent for pegging meters that were supposed to be set with plenty of headroom.

Re:Depends on the bitrate

[Bengie]

For me, MP3 knocks out a lot of highs no matter the bitrate. Listening to most Jazz really brings out the flaws of MP3.

Re:Depends on the source

[femtobyte]

The trick you're playing on yourself here is:

x = [1 0 0 0 0 0 0 0]; % x is only defined on 8 samples over the interval. There are an infinite number of continuous signals that could be sampled this way.

Following your procedure through to y:
octave:5] y = ifft(Y);
octave:6] y
y =
      0.87500 0.12500 -0.12500 0.12500 -0.12500 0.12500 -0.12500 0.12500

so y is also defined at 8 sample points; as for x, there are an infinite number of curves that could fit these. One of these curves is the sum of frequencies indicated by Y. But what does fft(y,256); mean? From the Matlab documentation,

"Y = fft(X,n) returns the n-point DFT. fft(X) is equivalent to fft(X, n) where n is the size of X in the first nonsingleton dimension. If the length of X is less than n, X is padded with trailing zeros to length n."

So, now you have y defined in a larger window (y = 0.87500 0.12500 -0.12500 0.12500 -0.12500 0.12500 -0.12500 0.12500 0 0 0 0 0 .... 0). See my response above to another poster's question: when you enlarge the sampling window, you "create" a lot of possible "intermediate" frequencies that "don't exist" (i.e. are indistinguishable from sums of integral frequencies in the shorter window). By padding y with zeros to a larger window, you're looking at a *different* signal from the un-padded y alone; consequently, you need the "extra frequencies" that you ascribe to the "non-sharp-cutoff" to correctly describe the different "y+0,0,0,0,...,0" signal (which is distinct from y). But that doesn't mean the cutoff isn't perfect as defined on the original signal x->y. In fact, if you periodically *repeat* y (y->y,y,y...,y instead of y->y,0,0,0...) you'll see the "sharp cutoff" still applies since the periodic signal is still the sum of the original frequencies in y.