Hello all of you,
At first I would like to apologize that I'm writing without searching the whole forum for what I'm looking for. It is because I don't have much time until July, but I can't wait to know what I would like to... So if there is already a topic that feeds my needs, please just link it here or ignore me... Also If there will be something not clear in my speech, please tell me - I haven't been speaking/writing English for a long time, and I have little problems with grammar, sometimes with vocabulary... But let's go to the main topic.

Are songs compressed to MP3 with bitrate 320kbps really transparent? Can anyone hear the difference between mp3@320kbps and FLAC/WAV, or between mp3@320kbps and mp3@256kbps? Let's assume CBR compression to those bitrates using LAME. I heard some time ago a lot about that, some people say that it is possible, some of them say that it isn't. Personally, I have tested many music samples of different genres, compressed to mp3 with LAME (not the newest version but now I can't find out which one, supposely 3.97) with few constant bitrates (32, 64, 96, 128, 160, 192, 256, 320), in some special listening room (with professional speakers, some resonators and absorbers) at university where I am studying now. I compared all of them together with themselves and with samples compressed lossless (FLAC), using this ABX program: http://ff123.net/abchr/abchr.html .
I couldn't hear the difference between mp3@128 and mp3@160 in most cases, also mp3@96 sounded ok, so I decided to leave 320 and FLAC. I'm 20 years old and I have been listening to the music using earphones just for two years, now I'm not, I also don't go to parties, so I reject possibility that my hearing is physically damaged more than average person's hearing. It was possible that music selected by me was not good for this test - mainly drum 'n' bass, trance, IDM, some rock and classical. Also the samples were only 5 seconds long because I thought that if they were longer my ears would become accustomed and it would be ok to listen that.
I also made a second session in the same room, with the same LAME and bitrates, but using only one song suggested by my teacher (made by Jacques Loussier Trio). Using the same utility I could hear the differences between mp3@128 and mp3@320 easily, between mp3@160 and mp3@320 hardly, I gave up at mp3@192 vs mp3@320. Way better than in the first test, but it's still way worse than what I have heard about some people, that hear the difference between mp3@320 and FLAC.

I heard also that people specialised in hearing the distortion of samples have 5% chance to hear the difference between those high bitrates (256/320) or one of them and FLAC. I don't know is that true, I want to have some evidence to know who is right. I found Hydgogenaudio Knowledgebase where it says:

"-V0 (~245 kbps), -V1 (~225 kbps), -V2 (~190 kbps) or -V3 (~175 kbps) are recommended. These settings will normally produce transparent encoding (transparent = most people can't distinguish the MP3 from the original in an ABX blind test). Audible differences between these presets exist, but are rare."

Do you have some reports from some researches about that? It's the most important thing in this post - my request for them. I would like to rely on good performed research than on people's subiective opinions. I know that even mp3@320 has worse bitrate than FLAC, but damn! I mean audible differences!

Uh... I hope that I wrote everything important. If no - correct me;).

Regards,
BioZ

That's the whole point of ABX testing for yourself. It is subjective. You need to use what sounds transparent to you. Everybody's ears are different.

EDIT: What sounds perfect to you might have noticeable artifacts to someone else, and vice versa.

The following is a generalization:

Samples that can be ABXed at 320 kbps by people trained to hear artifacts are fairly rare.

Samples that can be ABXed at 320 kbps by people who are not trained to hear artifacts are (virtually) nonexistent.

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.

In terms of performed research, most of the transparency research about audio codecs was performed on these very forums and they suggest that you don't need to go that far beyond V5 to be transparent on 50% of samples. V2 is perfectly transparent for me on almost all samples. In fact, I've never once heard artefacting outside of a blind-test environment with modern LAME.

320 is the maximum bitrate for MP3. There are some places where it fails, such as the "Everything is Green" sample by Abfahrt Hinwil. I can ABX that quite trivially, even at 320kbps. I go to parties and potentially have damaged hearing. I haven't been tested. I'm not sure I wanna know.

320kbps MP3 is transparent almost all of the time. Even when it fails it doesn't fail particularly badly any longer. We've come a long way from the swishy, garbly MP3s that almost scared me off the format when I started.

Heard of killer samples?

Lossy encoding is limited (some codecs are better or worse than others, but since the signal is never the same than the original, there's always the possibility that something went wrong).

With that in mind, MP3 is a relatively old codec, and it has some unfixable flaws (because if you fix it, then it no longer is an MP3 file).

Killer samples help to find problems with concrete implementations, and in some cases, show these flaws that i am talking about.

So, by experiment, MP3 can be transparent at 320kbps (and *sometimes* even at 128, but that's for another time). But by design, MP3 is not guaranteed to be transparent even at its highest bitrate.

While most here would agree that most popular lossy codecs can be transparent for at least 90% of available music, it would maybe make sense to introduce a "risk of intransparency" measure.

After past experiences with killer samples, mainly at HA, the following vertices could probably be stated:

The risk is lower for AAC (Nero, Quicktime) than MP3 (LAME, Fraunhofer) at any bitrate.

The risk is higher for certain genres, especially metal and smaller for classical music.

The list could probably be extended, but that was the first that came to my mind.

I don't know if we can expect anything really scientific in this field in the future. Scientists often want to follow hot new topics, that are both interesting to work on look great in your list of publications. MP3 is kind of old in that regard for a long time now. AAC LC also has also reached a state where only asymptotical progress is left to work on. So don't expect a well funded, statistically sound study too soon. If we want something like that, the web community must organize it.

Try this sample. All other samples I've tried are transparent to me at -V5. This one was easy to ABX even at 320kbps.

I read all of your answers, thanks for them;).

At first:

Uh, I misunderstood myself... I wanted to say that mp3@320 has worse frequency spectrum (is it called so in English?) than FLAC, but imho it doesn't mean that the music for our ears are also worse because of this.

I know that ABX (performed by one person) is subjective. According to my own ABX tests I decided to compress my whole music to VBR at ~210kbps (to have some "safety kilobits" but I think it's just for my psychical comfort;)).
I mainly listen to drum 'n' bass and within this mass of sounds it's hard to think about "wether there is an artifact or there isn't".
I also listen to almost all of music genres, but I just want to enjoy it, not to frustrate that it's a little bit worse than it could be if I had some super-dooper audio system.
As yet I don't work with music so I don't need uncompressed/lossless files. If I will need them - I'll get 'em;).

I feel weird when someone says that FLAC > MP3 on usual audio system, so it would be nice to have some kind of professional argument to show, that their opinion is as subjective as mine;).
If there isn't any as yet - ok, maybe we can do this in the future;). Personally I'm unable for now to gather enough people to make the statistics reliable.

Twostar - I will listen to it tomorrow. But... how many songs are such?
If you want to, look here what I listen to: http://www.lastfm.pl/user/buiosu (but I don't like this page too much - when I look at it I can't see any truly dominating genre; now I'm telling you that I listen mainly to drum 'n' bass and this sentence is in conflict with those stats; or maybe I don't listen mainly to dnb? gosh...)

Regards,
BioZ

Well, in reality lossy codecs is just a compressed interpretation of the lossless source, so they would - in theory - always lose, but they are optimized so *a minimum* of artefacts can be heard by the human ear.

If you fear artefacts, why not go with FLAC or another lossless codec? Personally I keep my FLAC collection at home and make LAME -v5 encodings for portable use. I have yet to find an artefact in my encodings, but then again the environment when i listen to it might not always be optimal.

It's actually pretty easy. Fire up foobar2000, run an ABX test. That shows them pretty conclusively that no, they can't actually hear the difference. I've done it lots. If you're of the gambling sort, bet some money on it. Easy money.

Afterwards I usually grab the eig sample and show them that I can ABX 320kbps MP3. Tends to blow some people's minds until they hear the artefacting.

Also bear in mind that someone may not be trained to hear mp3 artefacting and therefore not know what to "look" for.

I am not competent to say anything directly, but I'm summarising a lot of what I read here.

It seems there is no reason to be concerned about a slight overall loss of quality. At high bitrates, what goes wrong is mostly that a few specific sounds produce artefacts. If you read the reports on ABX tests, you often find people saying "I could finally tell the difference between the two samples by listening to the cymbal hit at 1.12 secs," or something like that.

The only possible general loss might be at very high frequencies (>15 kHz), if you have very good high frequency hearing, and if your music has lots of content in that range (typically cymbals). Harpsichords are also a problem.

Otherwise, something like VBR ~210 can be expected to be perfectly fine, except for the very few occasions when it isn't.

The only sure answer is to do your own tests, which you have: there is no reason to doubt your own findings because of what other people say. Perhaps 5% of them have really good hearing (exceptionally good), and have trained themselves to hear artefacts; the rest are suffering from illusions.

The more I read, the more sure I am that if you want better sound, the first thing to look at is speakers and the room. And the last thing, too.

Among 'audiophiles', one occasionally sees boasts of being able to routinely tell any mp3 from source, just by listening.

When queried further, these people typically have little experience with either modern encoding, or have not done ABX testing of their claim.

But there is certainly HA-approved evidence for distinguishing 320mp3s from source, on certain samples. In fact there's a thread active recently that includes examples of 'positive' ABX of a particular sample at 320kbps (which now could be classified as a 'killer' sample; I was able to ABX it at 192 VBR, which is usually transparent to me). It's by no means the only one. So 320 mp3 is certainly not universally transparent, i.e., to all people with all samples.

However, these samples are rare; and I do not ever recall seeing anyone here claiming they can ABX 320 mps from source as a matter of course, i.e., regardless of input, the way some 'audiophiles' claim. (If anyone did have this ability here, I'd predict them to be LAME developers or tuners.)

I seem to recollect someone making that claim (not universally, but regularly). He was, of course, called on it, and produced ABX logs. As I recollect, he listened to some especially apocalyptic variety of metal and was attending to the hi-hats. At which point, everyone told him to get involved in testing and tuning. An exception, but the response shows that 1) HA is not closed to the possibility of genuinely platinum-iridium ears 2) such feats are very, very rare indeed.

guruboolez has shown incredible skill at noticing artefacts with a specific subset of music as well. I don't know classical well enough to say precisely where, but he's shown some very compelling performance!

Most people, when told which source is which, will hear the higher bit rate source as better. When they are unable to replicate that in blind testing, they will consider it a flaw in the testing. ("Most" based on my experience with people and reading posts here and elsewhere. YMMV)

Thank you Canar for this homage :-)
I don't have that much experience with MP3 at 320 kbps. At such high bitrate I only remember hearing pre-echo artifacts with well known samples or similar situation I encountered with some CDs (castanets). With castanets.wav-like samples ABXing is pretty easy and no too hard even with 640 kbps freeformat mp3 (with some training maybe).

The answer to the original question mainly depends on the meaning of « transparency ». If someone mean absolute transparency (on every kind of music) the answer is clearly negative. The mp3 format is flawed and doesn't handle well sharp transients. There are other formats which I would naturally call transparent at 320 kbps (mpc, aac, vorbis). They may have their own problem samples but they are not structural ones and they could therefore be fixed (psymodel issues or things like that).
But if « transparency » mean perceptually and immediately identical listening experience, then yes, I would call MP3 at 320 kbps transparent. The existence of located problems doesn't ruin the whole picture, and unless someone mainly listen to sharp electronic, castanets, or experimental noise music he should be happy with MP3 encodings for 99.9% of his library.

The remaining 0.1% could neverthless disturb many people and damage the confidence they have in the whole performance of the format.