What Sounds Better, MP3 or Ogg?

I've never been able to make a clear decision on the subject. These days I rip all my CDs to MP3 at 160kbs which means about 80 megs for a longer album. With a 100g drive on order ($220. I remember paying more then that for .1% of that space) disk space isn't really the defining issue, but that doesn't mean I'm gonna rip everything at 300kbs just because I can. I'm curious what people think sounds better, and what bit rates they find to be acceptable for both casual listening, and more picky listening. Don't forget to mention what sort of equipment your listening on so we know where you are coming from.

Comments

Try VBR before you go to 300kbps

[Dast]

I don't know much about ogg, as I use mp3 for most of my music encoding. I've played around with various bit rates and finaly settled on what I felt was the best for me in terms of quality vs size.

I now encode all of my music at a variable bit rate 64-256kbps with lame. Lame 3.70 does a really good job of this and produces files (at least for the types of music I listen to) that sound very good. For the most part, they encode smaller than a 192kbps, as the average bit rate used is less. As a check, peeking at John Coletrane's Giant Steps, the average bit rate is right around 150. The bulk of my music averages between 160 and 192kbps.

The cool thing about vbr is that if the file needs more than that, is can use up to 256kbps to help make the harder to encode spots sound better. So I guess the worst case size you could get would be a song completely encoded at 256kbps (but I can't say that has ever happened).

I have a hard time telling these vbr 64-256kbps files apart from the orignal cd. Sometimes I can tell, but it is rare and difficult. However, IANAA (I am not an audiophile), so doing your own tests should help.

All of your standard tools should support vbr files. Xmms does a fine job. I did need to upgrade mpg123 to pre0.59s, however.

Anyway, consider vbr before you go straight to 300kbps.

How to do listening tests

[Dr. Spork]

If you really care which format sounds best and want your listening tests to be taken seriously, do them right. (I worry many people don't really care about the sound, and want to just take this topic as an opportunity to plug the format which they favor for political reasons.)

But if you do care about the actual sound, rip some tracks you like from different types of albums. Then, cut out one part of the .WAV file and encode it using different MP3 encoders and different bitrates. (Or, if you want to save time, use only LAME for MP3, because there's a near-consensus that it gets the best sound. Don't forget to try VBR.) Then encode it in OGG format, also at various bitrates.

Now, the important step:

Decode the OGGs/MP3s back to a .WAV file, and make sure you name your files so you know which is which. Now, ask your roommate to burn all these .WAV files on a CD in an order that will not be revealed to you. Also burn the WAV that never went through compression/decompression (see if you can identify it by sound). Now, get your best pair of headphones, go to your stereo with a pad of paper, play the tracks over and over, and take notes on which track sounds the best.

Only after you've decided which tracks sound the best can you ask your roommate which tracks were encoded with which method.

This is not hard to do, and absolutely necessary if you want anyone to take your opinion about encoder quality seriously.

spork

Define your fidelity level

[brink]

Given that mp3 and ogg are each lossy (there will be erasure of sound elements) to a certain extent, this question is almost sort of disingenuous. Ultimately the answer reduces to "What sounds good to you?" As the question is stated, it kind of sounds like you want to be sort of an audiophile, but not go all out (which I can relate to, trust me.)

What I mean by this is, are you trying to be as true to the original recording as possible, or do you just want decent sound? If the former, you're trying to approach hardcore high-end audio and you don't want ogg or mp3. If the latter, then just go by what your ears tell you -- from everything I've experienced, the two formats are virtually indistinguishable on a standard speaker setup.

Second, you're playing said file from a computer or some kind of mp3 player. How good are your speakers/headphones? Do they have the range, presence, crispness, etc. that you want? How good is your player's line out and D/A converter? How noisy is your sound card? Hell, how much RF interference does your computer produce or induce in the sound card? If you want to be really anal, what kind of cables are you using to run to the speakers (or stereo)?

Ultimately, since you know that you're going with something that's not going to be totally true to the original, you just have to go with what you think sounds good. You have to remember, not all ears are created equal. Go by what's good for you.

Having said all that (and at the risk of contradicting myself), with -specific- songs I've noticed a difference between encoding at 128k and, say, 192k. This is especially true when listening with quality headphones. Classical music in general or music like Orbital in specific seem to sound better to me at 192k. After 192, I personally can't tell a difference. Your mileage may vary. I've listened to two identical classical pieces, one compressed at 128k and one at 192k, over a friend's hifi stereo and there was a difference in hearable elements and sense of presence. Over my lofi stereo there's no discernable difference.

So, of course make sure you take this with however much salt you desire. It all comes down to what sounds good to you, and what kind of sound setup you're using. As the question was stated, it's difficult to give an accurate answer -- and of course, even a "correct" answer may not necessarily apply to you.

Including this one.

*blink* ye gods.

[Chris Johnson]

There _are_ no coincidences :D

I'm the guy who wrote up a 'sonogram encoder study' using a pathologically impossible waveform to encode, and then measuring how much different mp3 encoders fell apart, and in what ways. Like r3mix.net, I wound up supporting LAME, but with some explanations for what people find compelling about Blade and Fraunhofer, respectively.

You also should know that people have been pestering me to add Ogg comparisons for _ages_, even wanting to send me the files I couldn't encode myself on an OS 8.1 Mac.

Well, there have been some changes at Airwindows:

• new powermac to take on ADAT editing duties and run the quirky old transfer card I have
• OS 8.6
• Amadeus 2 v 3.2.3, which imports and exports Ogg- unsure quite what version- and Amadeus isn't free, but the deal is I _have_ bought it earlier and my registration number works on 3.2.3
• iTunes (more on this later)

And so, _yesterday_, I set about getting a preliminary look at Ogg Vorbis using sonogram analysis on my Encoder Hell test sound- put in half a day on it, and updated my site to include the new information. And today, guess what turns up on Slashdot? Spooky.

Now, I need to emphasise that the process wasn't exactly the same as last time- I had to include some 'control' sonograms using the same mp3s that I used last time (Frau 128 and Blade 320, strong but idiosyncratic performers of known characteristics) for comparison. It's preliminary, and I don't want to immediately go into a complete shootout again because (a) it's such an undertaking and (b) I'm not at all sure I'm using a current Ogg version here. That said...

Here is the result of this early look at Ogg Vorbis, and I think I managed to sort of exactly what Ogg is relative to mp3. Quotes from the final report:

"Conclusion: Ogg Vorbis, at least the version I tested, is not wildly superior to mp3. Used at bit rates under 192K it tries much harder to encode real high-frequency data, but on some sounds such as a tone sweep its sophistication backfires, producing artifacts that show up plainly in the sonograms."
"However, used at higher bit rates it strikes a very clever balance, managing to pull together the best qualities of wildly different mp3 encoders into a single sonic presentation. Again, it behaves similarly to the very impressive BladeEnc in tonal purity, but instead of the miserable transient behavior of BladeEnc, it mimics the overstated transient behavior of Fraunhofer. This could easily be seen as best of both worlds."

That is, to my mind, a pretty strong endorsement, requiring only that high bit rates be used (as is intended) As such, I think Ogg will only become more relevant as bandwidth and storage space inevitably expand. It also is, in my professional opinion, very well positioned to keep mp3 in check- mp3 can only maintain its dominance by not getting carried away with licensing and IP abuses, because Ogg is sonically superior enough to be able to take over _if_ given the opportunity of a situation involving harsh mp3 licensing, given widespread use of higher bit rates rather than low ones. (This is why I dismiss WMA- it belongs to yesterday, an era of limited storage space and harsh licensing restrictions)

Now, about iTunes? I have some observations that I'd love to learn more about. Basically, I picked up iTunes because there's a patch making it possible to install on system 8.6, and I did that- only to be startled by a distinct difference in sound quality which I have the background to interpret. Briefly, it sounds like iTunes dithers its mp3 output to 16 bit, instead of truncating it.

A bit of background: any decoder, either mp3 or Ogg or whatever, is effectively synthesising a waveform from limited information. It's adding harmonics together to produce a linear PCM representation that's piped to the sound output hardware.

I suspect everyone making mp3 players has been simply truncating the waveform to 16 bit on the assumption that it's low quality anyway and doesn't matter... until iTunes... which has startlingly better dimensionality and depth than any other player I've heard.

However- there's no patent on the general concept of dithering. Some of the fancier ditherers and noise shaping algorithms are proprietary, but I happen to know many that are actually GPLed...

...because I write them. And that means that although I am not a Linux C coder- since the code and the algorithms for quadratic and primitive root residue dithers and indeterminate-order noise shaping are in the GPL sphere, the Linux world can have those technologies freely- and the proprietary world can't. Which may mean that Linux players (mp3 or Ogg) can fairly easily boast strikingly better sound quality than proprietary ones...

It's exciting to see the pieces of a truly superior free audio technology come together...