Endless debate with audiophile friends over whether “perfect” or “CD quality” audio can be achieved via MP3, regardless of bitrate. R3Mix.net offers the best collection I’ve seen of facts, data, explanation, tests, and observations on the nature of “perfect” digital sound as it relates to compression and commonly available encoders/decoders, etc. I do disagree with the author on some points – I’m still no fan of VBR, which he advocates, and his Myths page would probably send chills down the spine of most hardcore audiophiles. Still, lots to learn at the site. Their Quality page points to the largest real-world MP3 quality test I’ve heard of, involving 300 audiophiles and conducted by c’t magazin. Conclusion: Transparency (inability to distinguish between CD original and compressed MP3 versions on $15,000 worth of audio equipment) happens at around 256kbps. Wish there was a good English translation of the test page.

Thanks to David Huff for the pointer.

15 Replies to “R3Mix”

  1. The Myths part is great. Thanks for the link. Having been in a studio twice I can absolutely relate. We mixed the audio on their amazing speakers but cross checked a radio like system. If anything, then these settings are as close to what we wanted to achieve as possible.

  2. Yeah, that’s pretty rocky. I had already given Alta Vista translations a shot and decided not to post the link. Great in theory but just too hard to read. Amazing how the killer translation app still eludes.

  3. …his Myths page would probably send chills down the spine of most hardcore audiophiles.

    I should certainly hope so ;) Having been around some very tweaky, hardcore audiophiles since the mid-1970s, I’m convinced that the majority of their cherished beliefs are pure nonsense.

    On this topic, I found a refreshing alternative to “woo-woo gullible audiophile psychoacoustic lunacy” in a review of an interesting set of kit-built speakers over at Dan’s Data. It’s always a treat to read Dan’s articles and reviews – guy’s got a great sense of jumor!

  4. Heh… I meant chills as in chills of horror because the guy is so wrong on some of these points, not chills of horror because audiophiles will suddenly realize how wrong they’ve been all these years ;)

  5. Nothing fancy – I do 192kbps CBR. The biggest reason I don’t do VBR is because it throws off time calculations in most playback apps. Since there’s no way to calculate track length by multiplying the number of frames by the bitrate, the only track length possible is an estimate. This makes it impossible to burn an audio CD if the burning app is trying to determine whether there’s enough time left on the disc. It’s a lesser hassle that, for example, the time in a playlist or library in iTunes is inaccurate as soon as you have even one VBR track.

    VBRs advantages are greater at lower bitrates. If I’m doing 192CBR anyway, and the r3mix setting averages around 180, there’s pretty much no quality difference, but you can count on CBR more than you can on VBR.

  6. ok, so the line of “transparency” for 16/44 recordings is at 256k. is that an argument for compression when the industry standard (for “hardcore audiophiles” at least) is moving to 24 bit/192 KHz sampling rate? its like saying the carl’s junior $6 burger (that’s the name of it – the “six-dollar burger”) is indistinguishable from a denny’s burger. gawd, its still fucking disgusting kibble, speshully when you can go to barney’s (let alone chez panisse), know what i mean?

    anyone else reading this — *I* am one of the “audiophiles” shack refers to (tho i hate that term and do not refer to myself as one, but i can’t control the blogger in charge).

    and even tho they are both disgusting kibble, the uncompressed one is at least whole, and disk space is not usually an issue (except for those of you using ipods, etc, and then who cares? not gonna get quality out of em anyway.)

    btw, my only bitch — you say “$15K system” as if that means anything. a) i can put together a very trashy sounding $15k system. b) i can spend $15k just on WIRE, so its not like it’s so expensive that it would have to be good. c) i can also put together a system i’d be comfortable using for such a comparison for probably much less than that. it’s like talking about the performance of a $50k car. without context, it means little.

  7. System cost isn’t everything but it does mean something. It’s just a way of saying, “This test wasn’t conducted on a K-Mart boombox.” In fact, it’s a way of saying “on equipment far better than most people’s.” You say you can put together a $15k system that sounds like crap. I say you’d have a hard time convincing most people that the system you put together actually was crap.

    As for $15k worth of wire, all I can do is “rest my case” against audiophilia gone wild.

  8. “Since there’s no way to calculate track length by multiplying the number of frames by the bitrate, the only track length possible is an estimate. This makes it impossible to burn an audio CD if the burning app is trying to determine whether there’s enough time left on the disc.”

    This is the biggest bag of crap i’ve ever seen…

    Any burning software can determine the length of a VBR mp3… As any player can.. !

  9. Rofl — of course any burning software (I think you mean “encoder”) can determine the length. That’s a no-brainer, since the encoder has access to the raw data, and no estimation is necessary. However, a player playing a standard VBR file does not know how many bits are in any given frame ahead of time, so again, it can’t know how long the whole track is. What it can do is estimate by examining the first few hundred frames, and in most cases, this is what you get – an estimate.

    Sometimes these estimates are drastically wrong. Even in iTunes, I occassionally get long silences at the ends of songs, caused by incorrect track length calculation with VBR files. Granted, the problem used to be more prevalent than it is today.

    Now, there’s a qualification. Xing decided to deal with this problem by adding an extra header, called the Xing VBR Header:


    to VBR files. Since the encoder knows the track length, it can deposit this info in the file’s header. If a player also knows how to read the Xing VBR Header, it too will know the track length without guessing.

    So: If the file was encoded with an app that knows how to write the Xing VBR Header, and is subsequently read by an app that looks for same, there is no estimation. Otherwise there can only be estimation (how else would the app know the length, since it cannot be calculated, by the very definition of what a VBR file is!).

    Track length data for VBR files is also sometimes stored in ID3v2 tags.

    If you want to brush up on this stuff, my book on MP3 for O’Reilly is still available in some stores:


  10. I’m in the wrong forum :-) but I don’t know
    where to go. I understand acoustics in general
    but I am not an “audiophile” and I know very
    little about MP3.

    My task is to take spoken English from stereo CD tracks, and compress it by at least 49.87
    (track size divided by 49.87 = max allowed
    output size). It’s easy to get 44.13 (the resemblance to CD samp rate is pure coincidence)
    but when I try other settings, the size don’t make mathematical sense. It’s also easy to get
    88.sumpin but intelligibility is borderline and
    it leaves lots of free space. It sounds _great_ at 44 but won’t fit.

    Soliciting redirection to a better forum, or a
    little e-mail advice. Thanks.

  11. Wes, I don’t think you need to worry about the math. If all you need is 50% compression, you can use pretty much any MP3 encoder at 192kbps or lower. 50% is such a liberal requirement, you’ll be able to get away with high bitrates like 192. But if bandwidth is a consideration and it’s spoken word, I’d go lower — 128kbps or even 96.

Leave a Reply

Your email address will not be published. Required fields are marked *