AES Paper on SACD vs CD

Discussion in 'Location Recording' started by DavidSpearritt, Nov 2, 2007.

  1. DavidSpearritt

    DavidSpearritt Well-Known Member

    At last, a paper in a definitive journal. My AES membership is worth the price. Reading it now.
  2. DavidSpearritt

    DavidSpearritt Well-Known Member

    Some more details of the experiment here.
  3. Cucco

    Cucco Distinguished Member


    I want to see more details.
  4. MadTiger3000

    MadTiger3000 Active Member

    That is a good article.
  5. hueseph

    hueseph Well-Known Member

    Doesn't the 16bit down conversion nullify any of the benefits of a higher resolution recording? It doesn't seem like a logical comparison.
  6. DavidSpearritt

    DavidSpearritt Well-Known Member

    You need to read the AES paper, but they put a 44/16 A/D/A into the playback stream from an SACD player and no-one could tell the difference from hundreds of double blind tests with hundreds of subjects including pro audio folk.

    The methodology seems fine and it passed the AES review committee. Now it remains to be seen whether it can be reproduced by others duplicating the methodology.

    I love it, its cat amoung the pigeons time.
  7. Cucco

    Cucco Distinguished Member

    So...let me get this right (I haven't paid my AES dues in nearly 10 years...)

    They put a CD against an SACD, but in the SACD chain they had a 16/44.1 PCM converter?

    Or are you saying that they double blind tested the raw sound of a SACD (through DSD converters straight into playback) against the raw sound of CD?

    I've done my own blind tests with members of one of the orchestras I record (same exact signal going to both the computer and the DSD recorder) and 8 out of 10 of the musicians I selected chose the DSD recorder as the better sound.

    I'll see if any of them would be able/willing to come here on the boards and "testify."

    What makes me VERY suspect of this is, supposedly MANY people can hear minute differences in converter from brand "a" to brand "b." That's why Lavry Gold and DCS are in business. If this is the case, then they're also saying that there is absolutely no discernable difference in converters in general since clearly two different A/D chips and likely different analog circuitry would have been used!

    Something smells here!

    (But then again, it wouldn't be the first time I completely disagreed with findings published in the AES journal - one reason I haven't paid up in so long!)
  8. DavidSpearritt

    DavidSpearritt Well-Known Member

    Nothing smells. They put a 44/16 A/D/A into the playback stream from an SACD player. They referred to it as the 44/16 "bottleneck". It was switched in and out, double blind. At levels ranging from 85-92dBL replay, no-one could tell the difference with it in or out.
  9. Zilla

    Zilla Active Member

    Unless my logic is loose, it would seem this panel would also not be able to discern a live mix feed from its a/d/a "bottle-necked" version. As Cucco mentioned, they are not just listening to DSD vs 1644 formats. They are also hearing the analog circuitry of the SACD player, the ADC, and finally the DAC.

    That they were unable to differentiate between the equipment (much less the pcm conversions) seems rather fantastic.

    Having said that, I can't say I feel that SACD is a sonic quantum leap from a quality controlled RedBook CD.
  10. Costy

    Costy Guest

    That's maybe the bottom like to it...
  11. hughesmr

    hughesmr Guest

    There is one major flaw in reasoning in this paper that is blatantly laid out in the first paragraph. As a practicing statistician, I feel compelled to point it out:

    "They prove beyond a shadow of a doubt, with literally hundreds of double-blind listening tests at matched levels, conducted over a period of more than a year, that the two-channel analog output of a high-end SACD/DVD-A player undergoes no audible change when passed through a 16-bit/44.1-kHz A/D/A processor. That means there’s no audible difference between the original CD standard (“Red Book”) and 24-bit/192-kHz PCM or 1-bit/2.8442-MHz DSD."

    You cannot prove the absence of an effect in statistical inference (i.e. when sampling from a reference population....that includes you and me....)
  12. DavidSpearritt

    DavidSpearritt Well-Known Member

    That is not stated in the paper Michael. That is from the blog entry about the paper.
  13. Zilla

    Zilla Active Member

    David, as I don't seem to have access to the paper, would you mind spelling out their exact conclusion?
  14. hughesmr

    hughesmr Guest

    OK, thanks for the clarification. If so,The Audio Critic has a ways to go before they achieve their stated claim of "accountability in audio journalism".... I'd settle for a little accuracy in reporting!

    Still want to read the paper...
  15. DavidSpearritt

    DavidSpearritt Well-Known Member

    Exact conclusion reproduction is impossible for obvious reasons. The paper is readily available for a nominal fee from

    Their paraphrased conclusion is that using ABX double blind tests with 4 different hi-res audio players containing a switchable 44/16 bottleneck in the output stream, through 4 different systems, all replayed at pink noise calibrated 85dBL level, over 500 subjects, men, women, students of audio, age and backgrounds well represented, no better than chance was obtained for detection of the bottleneck.

    Some info here.
  16. JoeH

    JoeH Well-Known Member

    God, I love this. I've always felt that a good CD when done properly, with the best sample rate at the front end, & good dithering, etc. can sound fantastic. Never done a double-blind test with CDs and SACDs, but I'm glad someone has given this a little kick in the butt.

    I'd like to see a similar test done with AD converters. I'm guessing there's not a whole lot of difference among the few at the top teir, either.

    So very much subjective stuff out there, it's nice to see some of it reigned in once in a while.

    Card-Carrying AES Member.
  17. HiRez Highway

    Here are some great audio asylum posts on the subject, including several from the BAS papers authors, courtesy of Mr. Ted Smith.

    "There's a lot lacking in the paper as presented.

    Posted by Ted Smith (B) on December 29, 2007 at 00:14:08

    In Reply to: RE: interesting tests posted by jazz1 on December 28, 2007 at 23:29:51


    And the authors have shown some interesting biases in the past as well as when they discus the paper.

    link removed

    Here are some random comments:
    link removed

    link removed

    You can read more if you want over on Prop Heads. Just look for threads where the authors show up.

    -Ted "

    In another post Ted Writes:

    ( link removed )

    " Posted by Ted Smith (B) on January 02, 2008 at 16:07:12

    In Reply to: RE: claiming to show a person's bias means little to data derived from DBT. posted by Duilawyer on January 02, 2008 at 15:50:42


    The only reason bias was brought up was that when people expect a certain answer they don't necessarily take care to design a fair test that might invalidate their expectations. It's precisely when tests show something unexpected that real progress is made.

    In this case (among other problems) for a portion of the experiment (and not separated in the data) they used equipment that has a noise floor higher than that representable by Redbook: so one wouldn't expect to hear the additional resolution from hi-res.

    Their responses to questions also indicate that the care that most audiophiles might take wasn't really considered.

    When they found out that if you turned up the volume a little you could hear a difference between hi-res and hi-res trimmed back to CD resolution they changed equipment.

    Their bias also shows up in the claim that this test shows something about CD players when (at best) it may show something about comparing hi-res to lower res (with a particular algo), but it doesn't address at all problems with playing CDs with real players: e.g. error handling, jitter handling, filtering with any different algo than they tested...

    They even admitted that there is a difference in practice, but they attributed it to mastering differences with no supporting evidence. We know from long history here that there is little reason to believe that that is the biggest difference and we know of a lot of counterexamples: not the least of which are the many SACDs who's Redbook layer is mechanically derived from the DSD layer and hence are clearly mastered the same.

    -Ted "

    Thanks Ted. Well said!

    George :)

Share This Page