Slashdot Mirror
Can We Really Tell Lossless From MP3?
EddieSpinola writes "Everyone knows that lossless codecs like FLAC produce better sounding music than lossy codecs like MP3. Well that's the theory anyway. The reality is that most of us can't tell the difference between MP3 and FLAC. In this quick and dirty test, a worrying preponderance of subjects rated the MP3 encodes higher than the FLAC files. Very interesting, if slightly disturbing reading!" Visiting with adblock and flashblock is highly recommended, lest you be blinded. The article is spread over 6 pages and there is no print version.
Most people greatly overestimate how well they can hear these differences, but the never actually try it in ABX testing. I tried it years ago and I can't hear a difference between most codecs at reasonable bitrates and unencoded originals.
Here is an old classic from the Hydrogenaudio forums, from someone would bought expensive head phones and set up ABX testing. He was very shocked when he couldn't even tell the difference between FLAC and Vorbis at 64kb/s.
ABX Just Destroyed My Ego, My perception of my bitrate needs was greatly inflated.
MPEG 1 layer 3 (MP3) encoding was designed as a 'perceptual encoding' algorithm where less "effort" (fewer bits) is given to signals that fall below below a threshold based on the other signals present. For example, a quiet tone close in frequency to a loud tone cannot be heard by the human ear, so no effort needs to be expended on reproducing it. All we're debating is whether the engineering behind this is sufficient. Certainly at lower encoding rates the distortion characteristics get very weird, though, and not at all like degraded quantization noise or analog distortion (Try it for yourself...) A few years back I decided to perform a little test one time to see how 192kbps MP3s performed. A self-avowed audiophile friend of mine lent me a copy of one of his favorite "reference" recordings (a Diana Krall jazz CD), and decided to give him a little test. I ripped his 'reference' song to .wav, encoded to 192kbps MP3, decoded the MP3 back to a second .wav file and burned a new CD for him.
He couldn't tell the difference much at all, and actually thought the one that had been through the processing sounded a little better. I couldn't tell any real difference on my studio monitors either.
MP3 is certainly good enough, at least at 192kbps, for portable use and on any 'normal' home system. I'd be interested to hear of any other opinions from similar tests.
I've been exposed to people who write audio codecs for a living. They can tell because they've become sensitive to the artifacts present in MP3s. They also can pick up problems with CD's that haven't been dithered properly. They can easily pick out MP3 even at 320kbps. These are specialists. But even in this study there was one individual who had a high success rate.
At 192K and a good pair of headphones with good material I think most people could learn pretty quickly to pick up the difference - loss of stereo image at higher frequencies is pretty easy to pick up.
There are also studies available that point out the advantages of high bit rate recordings - these enable the use of sophisticated filters that eliminate some of the issues present with CD sound. If you are interested and have a mathematical bent, look up the work of Meridian's Peter Craven. Again the differences can be detected by specialists. I'm old enough so that my ears are not good enough to pick up these improvements.
I rip to FLAC and convert for my portables because of these factors.
If you want to try some testing yourself visit hydrogenaudio. They have apps set up to do abx comparisons so you can test yourself.
One problem is the simple A/B and asking which people like better. Well that is fine if you are doing something like testing two compression formats to see which has a sound people prefer. That is not fine if the question can people tell the difference between compressed and uncompressed music. For that you need an ABX test. X is a reference uncompressed sample, A and B are randomized such that one is uncompressed, one is not. People are then asked to identify the one that is the same as X. A test like that lets you tell if people can hear a difference, regardless of if they like it or not.
Also there is another angle to why people might choose to use uncompressed music and that is if there is any additional processing (like equalization) planned for later. Psychoacoustic compression schemes can have problems when processed later. Reason being that they do rely on things like masking, in that because X is happening, we can't hear Y. However when the balance of the sound is altered, well then that isn't necessarily the case anymore.
How important is that? Probably not very in a lot of cases. However how important is storage space? Last I checked 1TB was under $100. Storage is cheap. There's not really a need to milk every last bit out of a file. FLAC'd discs are in the realm of 300MB for a full CD. Big deal. I'm got space to spare, so why not go lossless?
What it really comes down to is what is "good enough" really depends on the situation. Depends on the music (some kinds cause more trouble for encoders), the listener, the environment, storage constraints and so on. I mean 64k is good enough to recognize the music. A 64k AAC or WMA is fine, FM radio quality maybe, and even a 64k MP3 is listenable. Is there distortion over what was on the CD? Sure, but maybe it is good enough in some situations (like say you need to be able to transmit stereo audio on a single DS-0 channel).
I really don't like these tests that try to give the one magic rate is that is good enough for all situations. Especially when they use bad testing methodology.
Personally, I'm a fan of lossless compression because then there's just not any additional errors. I've got the space so why not eliminate potential problems?
"I think that the choice of playback software or drivers can also affect fidelity of the sound."
Indeed, pick any two media players that don't use the same decoder, they will both sound slightly different. Plus most default to 16bit, when the media might be 24bit (including mp3), or 32bit for FLAC, and others.
Driver implementation matters as well, I use the same drivers for my audio card for Windows XP (which the drivers were designed for), and Win7, but they both sound different, Win7 is slightly softer sounding, in XP it sounds a bit more over-driven, can't really compare between Linux distros as they almost never use the same drivers, and I usually don't bother to play with it, if it has sound, that's good enough.
I think any "optimized for metal" sound in Linux might be due to a more direct interaction with the hardware, which is great if you have hardware to support it, not if it's generic.
A $70 pair of SHURE earbuds has made all the difference in how I listen to music
Sorry, but no earbuds are worth $70. They are simply evil and will destroy your hearing. Get yourself some proper earphones, something that cups the ear and ventilates.
One word for you: Grado. Highly recommended and reviewed at audiophile places. The lower end models price out at about the same as those earbuds, but with a superior sound. I recommend the SR80 - probably the best headphones in the price range. See Grado Labs - a small company in Brooklyn that still makes their stuff in the US.
Disclaimer: I have no fiscal interest in Grado -- I just like their gear. I have owned a set of SR60 for over a decade, and I use my 4 year old SR80's at work. The SR-225 headphones I got for Christmas last year from my wife never leave the house - they are that good (And a bit pricey). These are the best sounding headsets I've ever owned or used.
Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo! http://goo.gl/J9bkO
Forgive me, but that sounds suspiciously like the kind of excuses I always hear from the linux fanbois. I've done pro multitrack audio recording on a Mac since 1987 (and been a linux sysadmin since 2003), and have tried *many* distros with a variety of high quality "supported" soundcards, but have never managed to get as good a sound out of alsa -- not to mention what an unreliable nightmare the whole jackd/alsa trainwreck is to deal with. Recently I started experimenting with FreeBSD, and the difference is enormous. FreeBSD and OSS, not linux and alsa, are finally rescuing me from bondage to the Mac platform. I give up on linux. It's great for servers, casual desktop users, and doing rescue operations on borked systems, but it just completely sucks for pro multimedia.
--
Posting AC because I'm sick of all the know-nothing fanbois who want to argue with me everytime I try to talk about this.
Most modern tape equipment has pretty much little to no track bleed/crosstalk (if properly maintained). There are way more variables that affect a tape that don't exist in the digital realm (speed, bias, tape material, wow and flutter, etc).
I think you're thinking of tape saturation, which is something entirely different.
I think you're spot on with that guess. For example, Red Hot Chili Pepper's cd release of Stadium Arcadium have been especially critisized for being too compressed (a result of the loudness war. Someone at hometheaterforum.com forum created a comparison between the CD and the LP (which had a much better mastering) release of the album, where you can clearly see the difference.
Now, the norm for most music released now is to mangle it in that way. And the audience is used to hear it that way too. So mp3 compression adding more artifacts to it and removing tones, thus mangling the music further, might sound "better" for a lot of the audience, because that's what they're trained to hear.
It's The Golden Rule: "He who has the gold makes the rules."
Eliminate the stuff which most of us can't hear?
I am very small, utmostly microscopic.
What's it to you if I think I can hear the difference when you think I can't. Maybe I can hear the difference, maybe I can't. It's a personal thing and if I preffer listening to flac or 'pure' music or iof I want to hear crap played from out of a tin can, it really shouldn't make diddly squate difference to you! SO GET OUT OF MY EAR SPACE!
On another train of thought...
If you want to manipulate the music and say put it in another format, up the bass whatever and save it again, you really need to work with the lossless formats. Then if you must lower it to MP3 to save space, cheep hard drives makes that need a little less. Of course, on your IPOD or other little device with your cheasy headphones, you might as well go to some quality 4 bit recording format and really spave space. You won't hear the difference once the earbuds are done with it.
-- Many men would appreciate a woman's mind more if they could fondle it
The problem with doing listening tests with the "average joe" is that the average joe has no idea what to listen for. They have "uneducated ears" so to speak. For a vast majority of today's listeners, if the sound track doesn't have an overbearing bass track and squeaky highs, to them it "doesn't sound good". They have no clue what a quality recording sounds like. If they even hear nuanced sounds, like the quiet echo of the symphonic hall's acoustics (in the case of such a recording), they don't know what it is and they don't like it. I have even noticed that for some people, if the music recording is not over saturated (loud and distorted) they don't like it.
Lossless, and even better, uncompressed audio is and always will sound better than lossy, you just have to know what you're listening for.
A large number of people rating MP3 higher than FLAC would then suggest that a large number of people chose randomly...
Please turn in your random number generator at the door.
In a sample size this small, you could easily see 75/25 splits without being able to determine that the results are not random. Remember that a random set is random from one through infinity. There's a lot of opportunity for 'luck' between those digits when you're only looking at a small portion of it.
Since they are testing people's perceptions this is in part a psychological test. You cannot conduct perceptual tests directly because perception is affected by the conscious mind. Thus if you asked people "do you hear a difference," you are likely to get many false positives since you are predisposing people to seek a difference. Instead you ask people which one sounds better.
"even if they can't tell the difference (something impossible to determine from this design) then they are simply guessing or picking one arbitrarily, and there is no way to determine if or when this occurred."
Actually there is a way to tell if this occurred--you compare the data set to what would result from pure chance, and look for statistically significant differences. If everyone is guessing then in the aggregate the experimental result should match pure chance (50% say one sounds better, 50% say the other sounds better). If a statistically significant percentage say one sounds better than the other, then you have proof that it is possible for some people to detect the differences.
Build a man a fire, he's warm for one night. Set him on fire, and he's warm for the rest of his life.
It's a fine test design...
If you have two identical pieces of music and you require people to rank them in order of preference, then the results will necessarily be perfectly random. This provides a built-in calibration.
Conversely, if the results are not random, then the people could necessarily tell differences in them.
Imagine if you merely asked people to say whether they perceived a difference but without asking them which one they prefer. Such a design would have no built-in "calibration", in the sense that it has no objective way of signalling when two pieces are identical.
ANYONE who thinks information recorded in tiny wiggles in groves and played through a bunch of springs (stylus, cartridge coils, tonearm, not to mention the non-trivial compliance of the record itself) and then amplified by two-three orders of magnitude is a more accurate representation than a full digital string (almost independent of bit rate) is deluding themselves.
You have no grasp whatever how a turntable works. The "tiny grooves and wiggles" are a very precice analog of the sound waveforms themselves. They are so accurate that in the early 1970s they developed "quadrophonics", a four channel system that had the rear channels modulated with a 40 khz tone in the front channels.
Meanwhile, a 15 khz tone on a CD has three samples per crest. With three samples there is no way to diffrentiate between a sine wave, a square wave, or a sawtooth wave; all will sound exactly the same.
Free Martian Whores!
I use Sennheiser headphones myself--not because I'm an audio snob who thinks you MUST have Sennheisers or you're listening to crap--but because my headphones tend to take a lot of abuse and Sennheiser is the only company that sells a complete line of replacement parts for their headphones (which can be easily installed by a consumer). I got a little tired of snapping one wire or bracket on a set of expensive headphones and having to replace the whole set.
SJW: Someone who has run out of real oppression, and has to fake it.
No, that's not quite right with wines either.
It used to be that anybody could buy a good wine -- if he was wiling to shell out the money. Any fool could walk into a reputable wine store with $100 and walk out with a very good bottle of wine. The expert was somebody who could walk out with a good bottle of wine after spending $15.
Things have changed. Vintners are very scientifically skilled at producing very consistent, reasonably good wine no matter what the year's growing conditions were like. As a result it's quite easy to find a pretty good wine for under $10. For example, it is not uncommon to find an Australian Shiraz priced at $8-$10 for a 750ml bottle. You don't have to look at the label or the year to know what you're going to get. These wines serve admirably in the role of vin ordinaire. They have a pleasing berry or curranty nose; their strong acidity and peppery flavor stands up well to food. Their thin finish, in which the flavor seems to evaporate on the tongue, hardly matters at this price and when taken with food. I am quite fond of these wines, but never bother with reading the label. I just fish them out of the cheap bin when I need a red to go with dinner.
So why pay $40,50, or $100 for a bottle of wine? You pay for the the unpredictability of artisanal methods. You probably won't get a *bad* wine, unless the bottle has been mishandled in some way. You might need to keep the bottle for a few years for secondary fermentation and slow chemical reactions to break down long chain alcohols and other compounds. I'm a mead maker, and when mead gets to its final specific gravity (net sugar/alcohol content), it tastes like paint thinner. It takes another year to be drinkable, and two beyond that to be something you'd *want* to drink.
Every decent $10 bottle of wine is as like it's peers (for the same grape and general region) as one affordably priced Japanese sedan resembles every other car in that category. Of course, rarity per se does drive up the price of "fine" wines, but fine wines as a general category are not rare -- your local wine store is full of them. It's just that any bottle is taken from a small batch produced from grapes on a single estate -- factors that lead to both high price and unpredictability. You can't integrate the chemistry of grapes taken from over an entire region, you've got to work with what you've got in a few hundred acres.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
and Sennheiser is the only company that sells a complete line of replacement parts
Exactly. What happen to me is that the wire on the inside of the cable goes out, and I lose a channel.... it's happened on almost every set of headphones I have owned. When I finally purchased a pair of Sennheiser headphones, I just had to purchase a $8 cable, rather than an entire new set of headphones.
I empathize with you, and have a suggestion. For most venues, the music has to be loud so that everyone can hear it, including the people in the back. This means that those expensive front row seats (or elbowing your way up in a small venue) doesn't get you better sound; it just gets you a better view. And yes, that means that usually the music is way too loud to actually just listen to.
My suggestion is to look into foam earplugs. You can get a pack of Hearos earplugs at Target for very cheap. They're these little foam guys that you roll between your fingers to make them fit in your ear, and they then expand in your ear canal gently to [b]reduce[/b] sound, not block it.
The first time I tried them at a show it was incredible. The wall of bass and noise that I was expecting was replaced with actual music! And when I left, I took them out and my ears were perfectly comfortable. I became a convert with that show, and now always have some with me if I go to see a band.
Alternatively, small bands and local venues tend to have more reasonable volumes. But that's not always the case, and I've definitely noticed shows where the starting band was a reasonable volume and each group got progressively louder.