source comparisons - mp3 and CD

[zowie]

Quote:

Originally posted by Vertigo-1 If you know what to listen for, it's actually not very difficult to distinguish a 192 MP3 from a WAV file, even off a crappy soundcard. Using my Soundblaster Live, and Rotel integrated driving a pair of Grado SR-225s, the WAV file contains an airier sound...a major key component that high end CDPs have over low end CDPs, and a key component that good CD playback has over MP3s. The MP3 of the same song sounds flat, and lacks any sort of 3 dimensionality, or depth. Lifeless is a good way to put it. The average person probably wouldn't know how to listen for this difference...the common reason why some people think all CDPs sound the same. They're out looking for tonal differences, when differences actually lie elsewhere such as in PRAT, soundstaging, resolution, or depth.

I agree with Vertigo. However I also agree that some exaggerate the defects in MP3 as well.

 

[Ricky Monk]

One common error here, I think, is in the definition of "quality".

To some, quality means "how good does it sound". To others, it means not only that but "how close to the original does it sound".

A CD original, and a 256Kbps mp3 may both sound excellent, and it may be tough or near impossible for someone to decide which is which, as both are artifact free and whatnot... but they will not sound identical.

 

[Ricky Monk]

Quote:

Originally posted by NewportGeek first off, CD's sound better than MP3's. Period.

I would say they reproduce the original more accurately; they do not necessarily sound better, at least not to 99.99% of the people out there. (talking high bitrate here, 384kbps)

Quote:

2: They use a mac to decode. Mac's interpolate to the full 16 bit 44.1 khz sample rate when decoding mp3s. This is essentially digital upsampling. While it theorectically doesn't provide any increase in sound quality, it does increase the audiability of certian sounds. This is just like oversampling a cd using a dac.

Bad logic.. it's not "upsampling". This is not a conversion between samplerates; it's decompression, there is a distinct difference. Nobody ever said it was an increase in sound quality; it's simply restoring the data format to the same specifications of the original (44.1khz, 16 bit).. just as lossy graphic compression like jpeg restores to an uncompressed bitmap with the same color depth & dimensions as the original bitmap.

Quote:

3: They use a tube amp for music playback. The delayed frequency response associated with good tube amps serves as a buffer to soften the coarse sound MP3's provide. Think about it: if you want to push you're brand-new digital format, why would you use an amp based on 80-year-old analog circutry? Simply because any respectable solid state amp would have exposed their MP3's for what they are; lossy.

Listening is listening. Nobody is claiming there is no loss of information, only that it SOUNDS just as good. If the listener has a valve fetsih, that's fine. You could say the same about any solid state amp

Quote:

I've spent alot of my own time looking into mp3 quality encoding. The bottom line is, bits are bits. Anyway you look at it, mp3's are a form of compression; and a lossy one at that.

Nobody ever claimed otherwise.


Quote:

No one ever says that when you compress and decompress and MP3 you get your original wav back. So you know at least some of you bits, and as it turns out, ALOT of bits (and consequently your fidelity) are going out the window.

Well now.. If you play for me a perfect sine wave, for an hour, I can compress that down into VERY few bits, and reproduce it PERFECTLY. Where did the lost bits go?

Let's not forget the end product here is what comes out of your favorite speakers/transducers/skull implants, not the actual bits. The bits are not important; the sound they reproduce is.
If you run the output of a 384Kbps mp3 through a scope, and compare it to the original, I'm willing to bet you find them *extremely* similar, a lot moreso than the 4/5ths of data you seem to suggest should be missing.


If we were reproducing full spectrum white noise, and comparing, then we would find any kind of lossy compression meaningless, if we are trying to reproduce every peak the same way. psycho-accoustically, though, we don't have to.. we just have to generate white noise, and the listener will hear the same thing.

 

[chadbang]

Sorry, I'm with Vertigo. It's relatively simply to hear that mp3s are no where as listenable as wav (full cd) files.

I had a friend who was a recording engineer for the BBC. He refused to listen to MP3s, saying they were horrific. I'm very fond of MP3s and the free music they have been bringing me. So I decided to prove to my friend that MP3s could sound fine by giving him a test.

I ripped ten tracks from some of my best recorded cds (rock, jazz, classical tracks). I turned them into MP3s at 320kps and then burned a cd with both the MP3s and the original .wav files side by side. My thought was I would test my friend to see how many of the MP3 files he could discern from the wav files.

Well, I never gave him the test. When I listened to disc, I had absolutely no problem hearing how much the mp3 encoded tracks suffered from compression. A "flattened" sound (some call this "air") has always been a failing of mp3s in my mind, and that sound was obvious on my test cd (ripped on a Mac using TAE and encoded with the Sound Jam Frauhofer encoder (I still prefer Sound Jam to itunes - better options and functionality.

I knew my friend would have scoffed at the mp3 encoded tracks, and he would have been right.

Golden ears? Well, yes, I've owned some Class A equipment over the years and although they may have been downgraded to silver over the years, like Vert, I can HEAR the different. Waveform read outs be damned.

 

[HD-5000]

I guess with all this news, its all the more reason for audiophiles to switch to either of the new high resolution audio formats. You may have read all the negative things I have said about SACD, but now I'm a believer. The sound quality of SACD is nothing short of astonishing!

 

[Pepzhez]

chadbang,

The Fraunhofer codec is simply an inferior encoder and will not produce a sound file that is worthy of comparison to the orginal WAV or AIFF. Compare a Fraunhofer mp3 to an equivalent LAME encoded file and you'll definitely hear the difference.

I don't believe that anyone here is making the claim that mp3 is perfect, because it isn't - and neither is CD or SACD or DVD-A, for that matter; nor do I think this should be an either/or question. The format can be rather useful and does have its place. I deal with CD, MD, DAT, vinyl and mp3 - and all have their specific purposes and uses.

Now then:

Quote:

the other problem i have with computer based systems is the machines. they are noisy with the fan and when absolute detail or silence is required, it dont work.

I have a great solution to this. When I "retired" my imac DVSE from active computing duties, I transformed it into a dedicated music server (and recorder) - digital USB out direct to my DAC and stereo setup and decked out with a 100 GB HD. Works without a hitch and the fanless (re: super quiet!) design of the imac is a true blessing. In fact, it's the only way I could stand to have such a system.

If you really want to compare the relative merits of mp3 vs. AIFF or WAV, I think a setup such as I'm describing is the only fair test. I DO believe that a LAME encoded mp3 at 256 kb/s is *usually* the sonic equivalent of the original AIFF file IF it is decoded on a Mac (I suggest Audion over iTunes) and played back through a high quality DAC (both my CD player and my Mac setup are decoded through a Van Alstine DAC). I will concede that some tracks - and relatively few, at that - do sound slightly different when compressed, but note that I say "slightly different" and not necessarily "worse". After extensive testing and comparison, I have seldom felt it necessary to even increase the data rate to 320 kb/s. (Additionally, I should note that my tests have indicated that CBR is superior to VBR. VBR - at least its implementation in LAME - is still too haphazard, or should I say "flaky".)

Loss of air, presence and dynamics is really not a problem with the proper use of a superior codec (LAME mp3 or ATRAC-R). Yes, there are indeed theoretically difficult or impossible to compress signals - no one denies that, but they are exceedingly rare in the real world.

In the end, no one - even the most golden-eared audiophile - will have a problem with a WELL ENCODED mp3 or MD, or will even notice a difference unless they very carefully compare the compressed track with the uncompressed original (and even then, it's invariably a toss-up).

If only MPEG-2 (DVD) video worked as well as LAME or ATRAC-R ...

 

[chadbang]

Pephez:
It's funny, but inbetween the time I posted and you posted, I decided to run another test. I hunted around and found a free program called "audacity" that would run LAME encoding for me on my Mac. I then ripped a track (Track 6 ("Hotel California") off the Eagles "Hell Freezes Over" CD - a well recording disc, even if I'm not that big of an Eagle fan) and encoded it with my Fraunhofer codec, LAME and left the wav file alone. I burnt them all on a cd to liesten through my main system. Yes, by far the Fraunhofer was the most inferior encoding. Dull and lifeless. And, yes, LAME is quite good! At 320kbs I would say nearly equal the cd. Now, Joe Walsh's vocal was decidedly less three dimensional, with a bit depth subtracted from the mid-bass region. Not horrific by any means and if I hadn't been doing an AB, it would have been quite easy to live with. (On the other hand I generally purchase my speakers for their ability to reproduce vocals and if MP3 were a speaker brand, I would have to pass!) But, yes, LAME is quite nice indeed. Now I need to get a dedicated LAME encoder for the mac. I wish I knew how to hack Sound Jam. But, yes, Pephez, I won't battle you that LAME is a very good codec and I'm sorry that I've encoded so many cds using Franhoefer!

 

[taoster]

now, why dont you go try FLAC (http://flac.sourceforge.net/) before you go and encode everything in LAME.

 

[williamgoody]

Taoster, are there any portable players that support FLAC?

 

[chadbang]

That's the question: Is it compatable? Now, nearly everyone can play an mp3 file. Is FLAC and MP3 format? Or is it it's own format like Shorten? I didn't see the word MP3 mentioned at that site. I want to create a file that no one will have a problem playing.

 

[taoster]

williamgoody, not that i know of.

chadbang, its like shorten so its not transportable though it has much support from many os formats. if portability is an issue, mp3 is still the easiest form... or wav of course

 

[Ricky Monk]

Flac is just like shorten or monkey audio.. lossless.

Not really an option for portables.

Really, if the issue is portables, if you aren't into a PCDP... your two main options for tiny portables are minidisc, and mp3. mp3, by far, has the most options.

Let's be serious. If we're talking about portable sources, you aren't going to get some super high quality sound. If you are listening for the difference between a good mp3 and the original source while sitting on the crowded subway, you've forgotten how to simply listen to a good tune on the ride home.

 

[PodMan]

chadbang, are you using any of the alt-preset (standard, extreme, insane) settings for LAME? They are tuned at the code level for maximum transparency.


the -r3mix setting from the site provided by the original poster of this thread is actually quite old and outdated compared to the superior alt-presets.

As good as LAME sounds now, I have read that one of the developers say that LAME can be improved by as much as 20% but it would require substantial code rewriting to do so. Right now the effort seems to be going into low bitrate portable settings as opposed to transparency that the original alt-presets targetted.

iRiver is developing ogg vorbis support for their mp3cd players but they are still in the process of getting it to decode in realtime on their hardware.

 

[halcyon]

I'd like to chime in with a couple of insights I've learned while participating in discussions similar to this and having read (and done small personal experiments in) psychoacoustics

1) Audibility of differences is the domain of experimental psychology, not mathematics, not computer science, not systems design. These things need to be tested in the domain of experimental perceptual sciences and whatever some statistician says, may or may not hold true for a single individual

2) People's absolute hearing levels vary from person to person a lot. What might be audible to one person, might not be audible to another

3) Source data (signal) matters a lot and so does intimate experience with that (original) source material. The more you have experience it, the more you discern it and

4) The ability to spot common playback artifacts (regardless of equipment used) varies a lot and is to a great degree a learned ability. Try the Excellent AES tutorial CD "Perceptual Audio Coders: What to listern for" for 6 months to personally experience this

5) You can prove that some difference exists, but you cannot prove that a difference does NOT exist. You can conclude that for certain signal, using certain compression, for a certain individual, using certain test equipment and certain number of certain type of tests, there differences might not be audible 99.99% (or some other statistical figure) of the time.

But you cannot conclude that the difference does not exist, for anyone, for any signal and always. That is beyond science and one neeeds to go back to the books on philosophy of science, if one thinks otherwise.

So, people, relax.

Even if somebody concludes that he/she cannot hear the difference between an original and a lossy perceptual encoder, using his/her gear while listening to certain type of material, then the only things this test proves is exactly that

It doesn't say anything conclusive about how you hear or how anybody else hears.

Even a test with 1000 participants and carefully done statistical analysis can only concludes some rough means for an average set up, average listener, average test signal and certain type of test setup (and familiarity with the test signal).

It is not *ever* a final say on what people actually hear.

So, don't get offended by what somebody hears or doesn't hear.

I'd be happy if I didn't hear any differences between 128 kbps mp3 and the original, as it would save me a lot of trouble, but alas I do and I have to live it. I cannot *unlearn* my hearing apparatus from hearing certain psychoacoustic compression scheme artifacts.

Best regards,
Halcyon

 

[JaZZ]

halcyon...

...good statement! I wholly agree. As you implied, there's a lot of motivation from the side of people who don't hear any differences even with low bitrates to presume self-suggestion with others who do. It's a very natural reaction which I know from myself, too, in a certain degree.

With my soundcard – the only source which provides the possibility to do such comparisons – and good headphones (HD 600 and SR-225) as well as good amps (MF X-Cans, Earmax Pro) my limit is at 320 kbps. I can't clearly differentiate MP3 from Wave at this rate. At 256 kbps there are few critical recordings which I can painfully distinguish. But 128 kbps are no question.

I expect that my main rig would resolve so much better that even 320 kbps could be distinguished. All those tests which seem to prove that even 128 kbps are as good as uncompressed files have been done with unfamiliar equipment in an unfamiliar environment. That's a very important factor! You need calm and a familiar environment and equipment to perceive the subtleties from which at least low-compression MP3s differ in comparison to Wave.

BTW: MP3 is great for portable use! A good MP3 Jukebox and there's nothing left to be desired – you have the whole or the best of your music collection with you.

JaZZ.