Slashdot Mirror
Audiophiles Test MP3, EPAC and MWMA
An anonymous reader wrote in to tell us that "Sound&Vision has tested three different "codecs" and compared the sound quality to a normal CD. The three are MP3 system, Lucent's EPAC, and Microsoft's Windows Media Audio V2. None could give full cd quality but MP3 was the over all winner."
Both Xing's and Lame's current VBR sucks. They both cause audiable distortion in some cases because the psyco model thinks it can go lower then it can. If you really want good VBR: Get the latest lame beta (3.25) and change #define RHXX to #define RH in loop.c, then change -X3 to have a max_noise of -1db (also in loop.c) and then encode with: lame -b128 -v -V0 -X3 -ms -k infile outfile This should produce ok results. But it would be better if you just waited for a few weeks for better VBR stuff. Also, keep in mind that encoding w/ VBR REALLY slows the encoding process. It's much simpler to always encode at 224 or 256.
What if it's a fact. Perhaps the encoding and compression of some music actually makes it "better" (remember that better in this case is defined as "less annoying for the majority of the population".
This reminds me of a similar article I read (probably about 12 years or so ago) in which a magazine ran double-blind test to determine the quality of T-120 vido tapes; IIRC the material was recorded on various manufacturer's tapes, and each recording was done three times (one for each speed, SP, LP, EP.)
The interesting thing (and the only reason I remember the article at all) was that the "regular joes" viewing the tapes frequently rated the EP recordings as giving the highest quality picture (EP always gives the lowest-quality picture.)
I think it's kind of like wine-tasting; if you get Joe Blow off the street to do a "blindfolded taste-test" with $100/bottle wines vs $10/bottle ones, you probably shouldn't place too much emphasis on the results if s/he picks the $10 bottle.
But the real question is, is the sound actually better (ie, closer to the source), or just more appealing to the ear?
I often hear people claim that analog media outperform digital for reasons like "a binary signal cannot possibly reproduce all the gradients in a musical tone", usually these are people that have not encountered Fourier Waveform Analysis and Nyquist's theorem (which states that if you want to exactly reproduce a signal of bandwidth H, you only need 2H samples per second).
Let's face it - digital mastering levels (24-bit, 96kHz) give a theretical S/N ratio of 144dB (using Shannon's equation), and faithful reproduction of sound up until 48kHz. You are telling me your ears are more accurate than that? Wait, let me put that in perspective - noise from a Harrier Jet engine at 1 metre is roughly 140dBW/m^2, and a silent room is usually around 20-30dbW/m^2. (For those curious, the figures for CDs are 96.3dB and 22kHz).
The phenomenon you describe is due to the fact that when tubes distort, the sound is nicer than that from a bipolar amplifier. This is because the distortion harmonics are greater on even harmonics rather than odd harmonics. For various psychoacoustic reasons, that sounds "better". MOSFET based amplifiers also have even harmonics when they distort, but are more difficult to get as linear as a tube. But they do make a top class bass amplifier.
For a soft sound, I like to mount my CD player on sorbethane. For a sharper sound, I use metallic spikes. Mounting it on a Rimu table I found gave a solid sound. My favourite is folded hundred dollar notes under each foot, which gives a very rich sound. And don't forget to circle the edges of the CD with a green pen to dull the internal reflections from the laser!
Err, if you look at the results of this test, in 7/19 musical selections, there is no statistically significant difference between the MP3 and CD versions of the same music. For two more songs, the difference is very small by any scale. For song but one, at least some of the audiophiles couldn't tell the difference between the MP3 and the CD.
I don't think the above results qualify as 'rather evident'. 'Almost no difference' is a much better description.
There are no consumer sound cards on the market that even come close to the output quality of a halfway decent component CD player. One of the main reasons for this is tons of electromagnetic noise inside the case, but also just because sound card manufacturers like to make money. I can pretty much guarantee if you've got a halfway decent stereo hooked up to your computer the weak link is the sound card.
That said, the best way to try to improve sound quality is make sure you've got one of the better cards on the market. Some good tests can be found here.
Also, for some Linux specific issues, the Audio-Quality-HOWTO is a good source.
All these comparements of "MP3" against other audio compression technologies are rather meaningless.
There is simple no "MP3" at all. MP3 is using a psychoacustic model for data reduction, and this model is not specified in the MP3 patents and therefore there are different models out there with varying results. I know of at least 6 models at this time:
- DIST10 The acustic model used by the ISO reference source. Said to be rather bad.
- BLADEENC Is basically the DIST10 model, but with few improvements and fixes.
- FRAUNHOFER Used by Producer, l3enc etc. Said to be one of the best.
- GPSYCHO GPL-model used by LAME. Apperently also quite good quality.
- XING/OLD The old Xing Encoder used this. Cuts the frequencies at 16 kHz. Increndibly fast compared to others, but bad quality.
- XING/NEW Apparently the new Xing Encoder (at least the linux version) use a new model, as there is a new switch for changing between cut at 16 kHz and not cut. To my tests the quality is ok.
So you see, testing just one MP3 encoder is not meaningfull. All these encoders have different qualities, different speeds. Some encoders have better sound at 128 kbps than other at 160 kbps or more. Use a bad encoder, and the result will be bad. Use a good encoder, and the difference to a CD will be heard only by trained people (these people who helped developing the psychoacustic models).
Additional every psychoacustic model will not match on all people. The human ear is just too complicated and different for a catch-all model. So even different persons may rate the encoders different in quality.
If i may offer a advise for MP3-Encoding: Use the new Xing-Encoder for Linux or LAME. Make use of variable Bitrate-Encoding. Fixed Bitrate-Encoding is bad, as the bitrate will always be to low at some very special pieces of the audio and very often just to high. Variable bitrate encoding tries to use the Bitrate just needed. I've made very good experiences using VBR and got smaller files which sounds better.
If you're a big audiophile, you're not going to be using MP3 anyway. MP3 is for a whole different group of pepole.
The issue with MP3 is its portability. The idea is that I can encode 10 or 11 of my CDs, burn it on a CD-R, and have a nice wide selection of music I can play on my desktop PC at work. Rather than haul around (and possibly scratch) a whole stack of CDs, I just have to carry around one CD-R that if I scratch up, I just have to burn a new copy from the files on my hard drive. I don't care about quality - heck, I listen to MP3s on a Sun with 8-bit audio. But it's great to be able to stick in one CD, fire up xaudio, and have 10 or 11 CDs to pick from.
Or, MP3 is nice because of convenience -- I can get my music quickly. Suppose at home, and I'm sunk deep into coding and don't want to be distracted. If I want music, I'd have to go to my CD collection, pull out a CD, walk over to the CD changer on my stereo, wait for it to turn on and spin up, and then play the CD. OR, I can just click over to the directory of MP3s on my PC, queue up a whole bunch, and have great music without even getting up from my chair.
Sure, MP3 is nice for downloading too, but unless you have a fast connection, who really wants to sit around and download a whole bunch of 4-meg files?