1. Register NOW and become part of this fantastic knowledge base forum! This message will go away once you have registered.

SRC-- what's in YOUR daw?

Discussion in 'Recording' started by Sonarerec, Sep 17, 2006.

  1. Sonarerec

    Sonarerec Guest

    Over on the proaudio list (where the degreed EEs and similar wonks hang out) there was a raging debate over Sampling Rate Conversion, and whether upsampling has made the numbers moot.

    One highly respected lister (who also designs and sells DACs) said, "As a rule, less calculations is better for quality. In this case, the 96 to 44.1 takes more calculations." The inference (based on other posts in the thread) is that it is better to record at 44.1 or 88.2 (integer multiple) if audio CD is the delivery mediium. It is for me, with the exception of orchestral projects for a label that also releases them in SACD and DVD-A. I should also add that they have taken material that was originally 16/44.1 and released it in SACD and DVD-A formats.

    The purpose of this thread is to ask what SR you record, how you get to 16/44.1, and if you did listening tests to determine your methodology?

  2. ghellquist

    ghellquist Member

    Personally I run at 44.1 through Lavry Blue. The results are plenty good enough for me.

    As for SRC - there seems to be a widespread myth that going from 88.2 to 44.1 is easier than going from 96 to 44.1. This myth is generally spread by people that has not designed and tested SRC algorithms. It certainly must be much easier is the lay persons guess.

    Wikipedia has a small article telling you a bit about how an SRC works. One important thing to notice is that if you write a good SRC it can translate between any rate of sample rates. There are even algorithms available that allows the ratios to be variable. This of course is a bit beyond the normal high-school level math.

    I should say that I have done some sample based algorithms, but no SRC-s. Frankly the part about sinc functions is a bit above my level of math.

    What I am trying to boil it all down to is that the only way to select sample rate is by measuring and listening to the actual equipment and algorithms you are planning to use. If you start apply theories about things that you do not know intimitely it is common for the conclusions to be wrong.

  3. TeddyBullard

    TeddyBullard Guest

    r8 brain Pro here...

  4. DavidSpearritt

    DavidSpearritt Well-Known Member

    Rich, I use 96K for all critical projects now, because some of them have unknown release media, ie some will be used for DVD-V release, which is 48 and of course everyone wants a CD.

    In my reading of SRC algorithms, 88 is never drop sampled to 44, it goes through the standard upsampling to 88 X 44 or 96 X 44 then re-interpolated. Correct me if you know of a specific example where 88 is drop sampled.
  5. Sonarerec

    Sonarerec Guest

    Gunnar-- if you read my post carefully you would see that the person suggesting the integer multiple "myth" IS someone who designs, builds, and sells ADCs. I will not mention his name as he was not asked about about quoting him, but it is a name any pro engineer would recognize and respect.

    I will email him and ask him to take part in this thread.

    David-- I am not sure about all SRC algorithms, but as I almost never record anything that is destined for a video format, I work at 44.1 and 88.2 integer multiples. It certainly will sound as good as other SRs and possibly be more efficient.

    Looks like the "compare and decide" poll is 1 did the test and the others have not.


  6. FifthCircle

    FifthCircle Well-Known Member

    R8brain Pro here... Darned good sounding SRC at a fantastic price.

  7. Sonarerec

    Sonarerec Guest

    I happen to know that Ben is a longtime Sequoia user, so did you do an A-B of a redbook-bounced file from Seq (at the slowest off-line setting) and compare to R8 Brain?

  8. FifthCircle

    FifthCircle Well-Known Member

    Not even close.. R8brain is a much cleaner sounding SRC- basically sounding much closer to the original rather than the coloration introduced by Sequoia's SRC...

    Now, SRC is one of the improvements in Version 9 of Sequoia. To be honest, though, I haven't really tested the new version.

  9. DavidSpearritt

    DavidSpearritt Well-Known Member

    I read Glenn Meadows mastering board but I can't post to it (lucky for all), as they still can't do this for GMail accounts. But I went through most of the long threads about testing the SRC in all the big DAW's as well as some of the offline SRC's like the R8BrainPro one.

    Wavelab 6's new Crystal resampler tested one of the best, I think they awarded it "most improved" in the end, over the very ordinary performance of the internal WL5 version. Before WL6 I didn't do much SRC'ing but used the free version of R8Brain and found it excellent.
  10. bassmutant

    bassmutant Guest

    Awesome, thanks for the link, been wondering what to set my logic projects at as I'm a recording noob, but a Math Major all the same. Kind of saddened to see logic pro not looking so hot. Once I get something else to compare side by side, I'll see if I notice. Good to see science dispelling myth. Any of these processes that depend on precision are going to benefit from more accurate processing algorithms as well as better floating point precision. Damn these sloppy binary floating point limitations, we need quantum computing. :)

    Also, here is a decent pdf link from Bias that has some graphed tests with information that seems relevant at a glance.
  11. Cucco

    Cucco Distinguished Member

    Yeah, David's right - there is no "easy way out" with SRC. Converting from 88.2 to 44.1 is just as intensive as going from 96 kHz to 44.1. You can't simply drop the doubled samples and get to 44.1.

    Granted, I don't write these algorithms, but I do have a team of scientists that do it for me and they all agree, there's no easy way out. (We have to alter SR for use of forensic purposes - capturing voice samples and transmitting them from hostile, network-constrained environments back to CONUS for comparison. The original file MUST be high quality, but for quick comparison, limited SR is perfectly acceptable. Bear in mind, compressed files are not admissable for what we do - it's considered an alteration of the file and therefore cannot be used. Therefore, we use a limited bandwidth version - seeing as how you only need 2x the bandwidth, not the maximum frequency - this little fact saves us a lot of bandwidth.)

    For my SRC, I find reconversion to be the best. I almost always feed out of the DA at the recorded rate into the analog realm and then back into the PC in 44.1. This is usually accompanied by some kind of processing (basic EQ or something - Crane Song - sounds OH so smooth even with no cuts or boosts.) Besides, the 44.1 on the Aurora actually sounds pretty darned good!

  12. RemyRAD

    RemyRAD Guest

    I think that Pulse Code Modulation sucks as a recording method! It is however hugely convenient and inexpensively executed. The truly good digital recording system is the Sony/Philips "DSD" (direct stream digital) system and that IS what the "SACD" CD is! Everything else is just PCM, which sounds like PCM, which sounds like PCM and all of the conversions require bad math. So either record at 44.1kHz PCM consistently and convert for video or start with something good like DSD. Mathematics only impress mathematicians. The rest of us want to hear something good.

    3 + 3=33??
    Ms. Remy Ann David
  13. Sonarerec

    Sonarerec Guest

    And how does your power amp cope with all the out-of-band noise with DSD?

  14. Cucco

    Cucco Distinguished Member

    It doesn't. That's the definition of "out-of-band."
  15. Sonarerec

    Sonarerec Guest

    Users of DSD have reported fried power amps from the ultrasonic noise that is "out-of-band" for our ears, but not so for the amps that are valiantly atttempting to reproduce it.

    Do you have a favorite term for the ultrasonic muck?

  16. ptr

    ptr Active Member

    Is there even an editor that works with DSD/DXD native...? All the ones I've heard of pops it in to PCM for editing... Much use for the beeing better then...?

  17. Cucco

    Cucco Distinguished Member

    Pyramix. (DXD - which, BTW, is PCM - 32 bit/384 kHz)

    I don't buy that. First, the noise is no greater in amplitude than the program material, just higher in frequency.

    Secondly, I've been listening to SACD (DSD) for years - almost immediately after it was introduced and my amps (ranging from Onkyo Receivers, HK power amps, Technics receivers, Rotel, B&K, Hafler, etc.) haven't ever issued one breath of a complaint.

    Besides, don't you think there would be some kind of class-action lawsuit against sony and philips if they put out a product which was knowingly causing damage to amplfiers and loudspeakers.

    I have to ask you now - have you ever actually listened to an SACD on something other than the crappy display at Circuit City? Most people who bash DSD usually haven't actually heard it in a real system.

    I'm with Remy 100% on this one.

    The HF noise is so high in nature that most components don't even see it and if it does happen to make it down the wire to the speaker, the actual frequency of the noise is naturally filtered out by the capacitance of the air between the tweeter's magnet and voice coil (a natural crossover/rolloff if you will).

  18. Sonarerec

    Sonarerec Guest

    I am neither bashing nor defending DSD, as it is a moot point for anyone whose delivery format is 44.1/16. Clients are sometimes curious about quad until they hear the cost, but DSD/DXD/SACD etc is meaningless to most, and except for the majors is dead in the water.

    Much of what is on the shelves started life as PCM-- that is certainly the case with my Naxos stuff that becomes SACD, but at least it began as 88.2 rather than 44.1, which is the case with more material than the consumer realizes.

    As to HF noise, you can tell Tony Faulkner it isn't there, as he fried a Levinson with it. He further stated:

    "Sonoma does not convert DSD to PCM for any main audio path activities. Nor does SADiE. Both Sonoma and SADiE create PCM files for the purpose of creating waveform profiles for the PC's display GUI, both Sonoma and SADiE create PCM files for the creation of the CD Red Book layer. That is where it stops. I am not completely up to date with Pyramix DSD, but certainly to begin with it could do no DSD processing (including edit crossfades) without conversion to/from 352.8k/32bit for the duration of the processing or crossfade. A retrograde step in my judgment, for a format targeting itself at being squeaky clean. Early unreleased prototypes of the first SADiE 2ch
    DSD workstation used pcm sidechains in the DSD processing for eq, but this is no longer the case in production models. Sonoma and SADiE use proprietary DSD chips for processing, and they work well. The only problem always in the background with DSD is the amount of high frequency error noise, and whatever you do it can get in the way - DSD workstations include assignable and configurable low pass filters to stop the noise getting out of hand.

    He has actually worked with DSD for release. Has anyone here who can share their experiences?

  19. Cucco

    Cucco Distinguished Member

    It seems that it would be impossible to diagnose this. How could one discern that HF caused an amp to blow? It would be far more likely (and I'm sure the engineers at Levinson - no longer Mark himself - would agree) that it is far more likely that improper load or spike/surge is the more likely cause.

    Unless something has changed within the past 3 months on both of these systems, that's not true. Both (based on the requirements of 1 bit science) MUST convert to PCM to perform crossfades. Additionally, very few plug-ins are capable of processing on DSD.

    This is why all of the editing that Telarc does is performed within the analog domain (converted from DSD to analog, processed, then reconverted back to DSD.)

    Sure. But I record in PCM first and then upconvert later. I've done one project recently with a local university whose CD (and SACD) will be released spring of 07 and I'm working with a choral ensemble right now for an SACD. I record in 176.4 and then send the individual tracks to the duplication house (in both cases, it is/will be Oasis.)

  20. Zilla

    Zilla Active Member

    Hardly impossible. The output of some early dsd dac's did output easily measurable vhf signals. In some units the output looked practically like an i-m distortion test. If the following amp has extremely extended freq. resp, the electronics very well may try (in futility) to reproduce it, possibly generating a lot of heat. This has probably been solved in contemporary units.

    It has been over a year since I have done any tests, but my experience was that the best pcm conversion was preferred over dsd. In our tests the dsd was further sonically from the analog source than the pcm. I always had the feeling that I was being micro-waved when listening to dsd. But the technology is very young. Its possible that many improvements have been implemented since that time.

Share This Page