DSD is totally worth the effort.

But you said "I find it affects headroom". "Headroom" is a bit of a nonsensical factor when we're talking about a finished 2 track master (you don't need to "leave" headroom, you can normalise to whatever you like and not have to worry about "unexpected peaks"). It sounds like you mean "dynamic range"...but sample rate doesn't affect dynamic range either!

The comment above about MP3s is a little misleading too - yes, they're significantly smaller but it's about perceptual coding, not just throwing away information randomly to get a smaller file size.

And apart from soundcloud, who uses 128kbps files any more?! 192kbps should be the minimum shouldn't it?

It's really difficult to compare, short of tracking the exact project to 2 rigs, and scientifically approaching the mix process, to work out exactly what the benefits of higher sample rates are when the end result is an MP3.

That said - I recorded a piano/vocal today (bonus acoustic track for an album re-release) and I did it at 96k...because...well, I could!

If I start at 88.2 or 96...or even 48K...and I bounce down to MP3 using the same compression...VS starting at 44.1K, I find that I can turn things up a little louder, and things feel less blurry. It's subtle, but yes I would say at that point you are affecting headroom as well as clarity.

If I bounce down to DSD the image is even more detailed and I can go louder again (slightly) without feeling like things are starting to crap out.

We're talking about a war of inches, but DSD was a real step up...just like printing masters to 88.2K was for me prior to that.

Also 128kps was standard last I checked though 192 is gaining a lot of ground fast. (Maybe I should catch up with you there and start prints at that resolution.) I tend to do 128kps and 320kps prints for clients after archiving.

agree again about the step up.

from 44.1 to 96 for me was "a real step up"

96 to 192 was kind of different to a larger degree slightly better i thought.

192 versus dsd128 well, we're back to "a real step up"

it just has clarity, depth etc etc.

based purely on what I've heard i just wouldn't feel i was being serious to my music recording at 44.1khz.

for me it goes,

dsd
tape
pcm96/192

in that order.

if i could use dsd for tracking then i would without even blinking. that gear sounds real to me compared to other stuff I've heard.

it's interesting that the article i linked to seems to suggest that the test people could accurately identify a difference between sample rates.

that certainly matches up with my experience.

*pedantry alert*

You mean 'word length', right?

But like I said..what are you comparing to? other mixes done at different sample rates? or the same song, recorded at a different sample rate? I'm not saying you're wrong, only that it's incredibly difficult to maintain perspective.

I still don't think "headroom" is the right word. You don't "hear" headroom. "Dynamics" maybe is a better term. I still don't think that's what's happening, but maybe there's something in the psychoacoustic coding that's saturating up top etc.

I have no DSD experience, so I can't comment there. I do struggle to see how it could be more accurate than "indistinguishable from board capture" as I find high end converters to be when I've tried them.

Get with the program! iTunes is all 192kbps now, I think bandcamp etc are that high if not higher. The only thing I'm aware of that uses 128kbps is Soundcloud.

I am talking about a series of tests using the same files, most of which were conducted on and off again over a period of 8 months in 2008. (Not counting DSD comparisons, which I started doing this fall.)


I have to write a longer reply for this one sorry, and flip it's order with one of your other quotes:

Before swapping to DSD recently for summing I used to record with two computers...one with a session at 88.2K to sum to. The other one at whatever the original session was tracked at. I would go out from one DA into another, usually nicer AD, and after capturing the 2 Track master at this rate I would have the absolute best sounding starting point I could afford. I'd read this lecture given by Bob Katz (LINK HERE) in which he talks about why 24Bit 44.1K can be improved, but never as much as 96K. (Also I'd had recently read his book.) Meanwhile, Andy Seagle, who runs the music engineering program at MCC in Mesa AZ had a couple of teachers under his employ who had spoken to me about the existence of remainders that can't be rounded out with odd number sample rates as you convert files during a bounce down. They had postulated that 44.1K and 88.2K couldn't be compressed as seamlessly as 48K and 96K without leaving these remainders, and they wanted to know if there was going to be sonic differences there. Since I was conducting similar tests already for my job at the time, we tried it out. We found that while higher sample rates up to 96K had better "depth", stereo image, and overall clarity, that there was also something we could hear in terms of an increase of apparent volume as you turned your master level down when you compared 44.1K to 48K, and 88.2K to 96K. (We also found that going up to 192 was different...but not that effective at the time. This was partially due to computer limitations in 2008.) Prior to conducting these tests each of us had read that numerical values which are divisible by 44.1 made for less errors during downsampling to 44.1K. (Taking this for granted as fact for years.) But...we had recently had another unrelated math experience involving remainders with processing in Pro Tools. This lead us to believe that remainders existed audibly after compression when bouncing, especially when bouncing down further to MP3.

To us this meant that these remainders were additive properties. (Think of it like digital cross-talk if that makes it easier. Though that's not totally accurate of a comparison to make.) I could make things feel louder at lower volumes with files which had a little less by way of resolution going on, but it was a trade. After doing a lot of personal tests with my own sessions from there I wound up settling on 88.2K being a compromise I could live with. (I was getting the most of both worlds this way.) Things weren't that different sonically from 96K, and I had a bit of that apparent volume thing going on where the mid range felt louder / more detailed even after bouncing down to 128kbps from that starting point.

Simultaneously I had noticed that when I had two rigs. One to do the processing of the playback, and the other to do all the critical processing of recording, that the strain of doing both on one computer had been lifted, and there was a slightly noticeable difference in headroom from there. Similar to what I had experienced recording through buses in pro tools and then exporting regions rather than doing real time bounces, only with even better results. So this is where I start equating headroom with the kinds of dynamics preservation you are discussing.


When it comes to DSD recording, some people say it's similar to summing to tape, and what I think they're trying to explain is something along the lines of what you're describing here, only at higher resolutions. When you skip the self-compressing you will experience with a DAW conducting playback AND a bounce down at the same time, the step up is in the noticeable difference in "life" when it comes to transient response. Depth, clarity, stereo image, etc...all of this is preserved, but it's essentially an improved system similar to my old two computer rig when I record to a unit with DSD capability. The final side-effect from this point is again, a small step up in measurable headroom. I can simply just go a little louder without things getting trashy. Since I'm going to be reviewing the new DSD capable MyTek soon I am excited to see where it stacks up against the Tascam DA3000.


This is a recent sea change, but that doesn't mean everyone has swapped over just yet. Still...it's probably time I rethink my approach and start bouncing to 192kbps. It's not like "Mastered for iTunes" hasn't been a looming shadow since they started in that direction, and even Youtube is improving it's sound quality potential lately. Since I'm building up equipment for a new place I haven't been working as many sessions since last June. You make a great point, and I think it's time I really think about it more seriously since it's not a major adjustment to make.

Hi, I'm not doubting that you heard real differences -- there are many reasons why convertors at different rates can sound different, especially with older convertors -- but it's unlikely that it had anything to do with sample rate conversion. The SRC process is the same, whether the target rate is a multiple of two or not; one is not mathematically more difficult for the computer than the other. Not sure what you mean by "remainders".

Pro Tools, by itself can't totally deal with prime number remainders, it can't round them down or up, instead it creates additive artifacts.

Say I have a fader and I clock it -0.3dB I will actually increase apparent volume due to this. If I meticulously adjust all settings on plugs and faders to include a prime number at the end of it, I create a lot of these weird "additive" properties. Your numbers can be positive or negative gain structures, but you need prime numbers at the end of each setting to push this further to a largely measurable degree.

This does NOT happen with other any other DAW, just Pro Tools systems. To me though it's something that can quickly be turned to your favor, since it helps with quick decision making when sweeping for settings. (Try it and see. Spend ten minutes with a session and start implementing prime numbers while listening back.)

Because we could repeatably prove this...we set out to see if there were other systems or instances where remainders would cause a similar result. (Mostly for fun. The other stuff we were working on was a bit overwhelming at the time.) I have found that at 88.2K I'm getting that simpler divisible quality Mr. Kats talks about, but when I then compress down to MP3 from there, it's possible that something similar is going on during the file compression process to a similar effect. At least, I'm hearing a similar end result.

I'm unclear what you mean by remainder here, and I don't understand what exactly needs to be rounded. Can you elaborate?

Are you saying that -0.3 dB is louder than -0.29 dB? No idea how this could be possible, but how are you determining this?

I use Pro Tools and often I end up with faders at -3 dB; never noticed anything unusual. What should I be listening for?

Sure. As long as a value for a setting ends with a prime number you can use this to your advantage to create additive sounds which will increase apparent volume in the mid-range using Pro Tools.



We conducted exhaustive listening tests...over the course of months on this. (It was a something done for fun since I was working 15 hour days at the time on something else in a sound lab, and we had access to just about every DAW but Reaper.) It affects mid-range frequencies causing anomalies with apparent volume. My advice is to sit with speakers you know really well and try it for yourself. Think like digital compression artifacts only they increase volume overall.


Copy a file. Put one copy at -0.3db leave the other a 0. A/B them. There should be a slight difference with more mid-range poking through in a focused way at -0.3dB. Vs 0. But don't stop there, as you incorporate plugs and high track counts with a mix, spend time leaving prime number remainders on all parameters on the side which is at -0.3dB. You should find that without touching the volume on your speakers that the -0.3dB setting feels louder and more present. Better focus and clarity. It will be "apparently" louder. Especially if you leave the exact same settings on the exact same plugs on the other channel, without the remainders.

if you move an EQ to 800Hz...make it 813 instead, etc. Or 800.07, etc. You can sweep between prime numbers for subtle differences, but for whatever reason as long it's not something that can be divided equally without remainders there are artifacts which accumulate. Tiny additive artifacts like this can really dictate decisions in a mix over time.

I wind up changing every setting in my mixes to have a prime number involved this way because it does in fact change things for the better, causing things to be a bit more 3D.

People tend to scoff at first, and then I show them, we'll sit in front of several monitors in a row, with the same A/B test. Or we'll go through a session they're working on and I'll adjust their settings by incorporating prime number remainders, and it's very nearly always for the better on virtually 99% of their settings. (There is the occasional sound which doesn't need that kind of buff in the mid-range to blend right, but the whole of the mix will tell you if that's the case near the end.)

It makes mixing in PT easier and faster for me. But other DAW's do not work like this from what I can tell. (More sweeping for sweet spots occurs because volume resolution is treated more naturally.)

With other DAW's the only conclusion we could come to was that remainders like this are rounded out. But in PT they're added back in. Much like an exact copy of a signal would double that signals volume, this was introducing tiny pieces of the signal to itself somehow, causing an increase of apparent volume via the mid range. If this is a side effect of something else...I'd like to know, or if the highs and the lows are really being affected then that would explain it (Though I think a more complicated hypothesis would have to cover what was happening in that instance), but to me it just feels like the Mids get an ever so slight nudge up.

If you want you can PM me to swap info. I can skype or call ya, and walk you through some quick tests if you need more that what is detailed here to go on.

The easiest source file to test with at first is a snare you're familiar with. Just an FYI.

I tried this and couldn't hear a difference. Of course that doesn't prove anything, but I'm trying to imagine what might cause such an effect and I can't come up with anything. What's particularly baffling is why prime numbers would matter. Besides the fact that neither 0.3 nor -0.3 is prime, every integer is either prime or a composite of primes, which means that every rational number (like 0.3) is a composite of primes: 0.3 = 3/10 = 3/(2*5).

To put it another way, any calculation the computer performs can be re-written such that every single number used in the calculation is prime. And the answer must be the same, either way. If this weren't the case, math would be broken and computers would be useless.

Rounding errors is a different, and very real issue in computer calculations. But with 32-bit float internal processing, the rounding errors are below the noise floor of a 24-bit signal. Effectively, they don't exist. So I'm at a complete loss to explain why you would be hearing such a thing.

I'm saying whatever the final number(s) in the sequence if it's a prime number by itself it causes rounding errors like you're describing. (Maybe my language isn't clear enough?)

So if you add a .03 to an EQ setting for example...on the last four versions of Pro Tools I've used this is will cause the mids to focus up ever so slightly. And so on, right down the line. If you make many changes this way on a channel with several plugs in to, the whole of the sound focuses up. You do this across a mix, and things start to really sound more defined without really moving far from your starting points with settings.

What are you listening back on?

Also you can sweep different prime numbers...say I take an RVOX for example...maybe -0.07 would be a better setting that -0.03 if I was trimming it's gain back just ever so slightly while compressing.

But if I was in Cubase...things wouldn't focus up that fast, I'd have to sweep the whole of it to find the right spot every time.

It seems we may have yet another one of those GS reality-disconnects... hope I'm wrong. But I can't figure out a way of interpreting HCYM's talk about prime numbers that makes any sense to me at all. I mean, so far, it seems like he's talking about decimal fractions, which, of course, are nothing like prime numbers.

Prime number - Wikipedia, the free encyclopedia

First thousand prime numbers

Picking a 3 or a 5 or a 7 is usually what I do because it's simple to tag on the end of a ballpark setting, easy to sweep between the three of them. Etc.

I'm sorry... why are you laughing?

What you are saying doesn't seem to make any sense, assuming normal definitions of the terms you are using.

Is that really all THAT funny to you?

I'd like to think you at least know the difference between a digit and a number...

By adding a prime number as a remainder to a setting you choose I notice a difference. If I keep adding these slight tweaks down the line it starts to really add up.

All I'm saying is give it a shot if you run a Pro Tools system. If you really don't hear anything maybe I'm crazy. But yes a full digit is different than a number (single symbol, etc.) But you know well and good that's not what I'm after with what I'm talking about. I'm using fairly simple language.

OK. I think I see what you're getting at.

If I understand, you don't actually mean a prime number. A prime number is a number, a value. You apparently mean a digit that would be a prime number if that digit -- standing alone -- represented a value, ie, the digits 2, 3, 5, and 7.

And you are then adding these digits to the end of a decimal fraction value in PT and you feel that is creating audible differences that you feel are not producing the expected results?

I'm afraid I wouldn't have any way of experimenting with that, myself (no PT). It is a bit perplexing how the correlation you suggest might arise but since I don't have the software in question or any real insight into its internal processing, I'll just back on out, as I don't think I can be helpful.

Correct on all counts!

I don't see how the last digit being prime can cause rounding errors. Here's an example: let's say I have a track and set the fader to -0.4 dB. What happens inside the DAW? It's basically two steps:

Step 1. Convert the gain from dB to linear:
Step 2. Multiply every amplitude in the signal bit stream by this gain factor of 0.954993. So if the amplitude is 0.435051, after the fader we get an amplitude of:
Now what happens when we set the fader to -0.3 dB?

Step 1. Convert from dB:
Step 2. Multiply every amplitude by the gain factor. Using the same example amplitude from Step 2 above, we get
I performed these calculations using IEEE 32-bit floats, so the numbers and rounding are representative of what they would look like in Pro Tools. (The 64-bit version of this is much the same, just with much larger numbers.) What possible reason could the last digit in the dB representation matter to the computer? It's just a multiplication of two numbers; rounding will happen regardless of whether the last digit is prime or not.

Finally, consider that the computer stores all of these numbers in binary representation, so the "last digit" being prime doesn't translate.

Cool. Sorry for the distraction and good luck on your effort. I'll try to check back in and see if any new info turns up.



PS to no one in general... To oversize images, add 'code quote' text boxes... heh

This is where I'm stumped too...but I can hear it.

System to system. I'm hearing it. Then I swap to any other DAW we've got around here, and...I don't get the same result at all.

My bad! I edited the code quotes to be smaller as soon as I noticed the oversized result. Hopefully they're manageable now; if not I'll surely nix 'em.

Edit: Looks like I wasn't fast enough: HCYM quoted my orig; think you could delete the code blocks out of my quote?

Confirmation bias - Wikipedia, the free encyclopedia

If you didn't do an a/b/x test then you didn't hear the difference. The human mind is an amazing thing, but it's not all that great at being unbiased by what we expect.

i have a confession to make.. for me it's the opposite! i've always had the impression stuff sounds better @ 16/44.1 when tracking but for some reason i use 24/48 anways.

left brain wins this one.

By moving things around, I was able to see everything. (Also, it's worth noting that many modern browsers let you ctrl- or option- + or - to graphically zoom the whole page. Some browsers let you control text and image zooms separately.)

Got your PM! I fixed it...sorry!

Confirmation bias and other cognitive distortions are difficult for humans to escape because a large part of our 'success' as a species is the ability to apply heuristic shortcuts to understanding the world 'well enough' to get through the immediate situation.


I recently was reintroduced to a peculiarly visceral perceptual illusion, I guess one might say.

I had made some changes in my rig and sat down at my heavily piano-weighted MIDI controller. When I played a few phrases I was profoundly alarmed because I had the distinct impression that the action had suddenly become sluggish... it felt physically hard to push down the keys. Since there are no external adjustments (it felt even across the 'board), no levers, mechanisms, etc, I was perplexed.

And then I remembered I'd had the same experience a year or two before, really beside myself because my keyboard's physical action suddenly felt like the thing was filled with slowly drying glue.

But -- and some of you probably saw this coming -- it was just that the velocity curve on my grand piano VI had got reset to the default, but I was used to a 'lighter touch' -- so it was simply sending lower velocities than I was used to, causing me, apparently to play harder but also -- for a good half minute or so -- convincing me there was something physically wrong with my keyboard controller. I could really feel it...

Anyhow, on about your business.

Thanks, and it was my bad!

as far as signal integrity goes this is pretty much spot on. should perhaps put analogue as in straight wire at the top though!