Yep, so he's saying that at higher levels, you're using more more bits to represent the data, and at lower resolutions, you're using less bits. This is what I thought he was saying in my original post. I think this is completely wrong and is not how digital works. I can't say I'm 100% sure, but I'm 95%. I'll gladly accept it if I'm wrong.

It's like people who say that 24 bit has more resolution that 16 bit. It doesn't. The resolution at a given level is the same, but the lower end is extended, so you can go even quieter before you hit the digital noisefloor. There is not a 'finer' resolution in 24 bit. It doesn't use more increments. The increments are the same. There are just more of them in total. These increments are the same amount over the entire recording level. They are not all squashed up at the 0dB region leaving only a few down lower.

Further more, once you introduce floating point, it doesn't matter where things are. You got a file recorded with a peak at 0dB and a noisefloor at say -120dB. If you aren't using floating point, save the file with a -60dB reduction and you now have a -60dB peak with a -120dB noisefloor. Bring it back to 0dB peak and your noisefloor moves up to -60dB. With floating point, that noisefloor moves with the peak down to -180dB and then back to -120dB. It's got some extra info in there simply providing an offset. At least that's how I THINK it works. Again, correct me if I'm completely wrong and I'll be wiser for the knowledge.



At the end of the day, if what he is saying turns out to be true, I really don't imagine anyone is going to hear the difference, assuming your gear is quality and not introducing other issues to cloud your test. I'll do a few tests when I get a chance and see how it sounds. I don't actually expect to get the same sound result but I'm guessing it'll be due to my equipment cheapness, but we'll see! I'll do maybe -3dB peak, -18dB and -60dB. Something like that should be interesting.

Ben is basing his argument on the fact that Stavrou's book was written in 2003 and a lot has changed since then. He is not taking that into account. There was some truth in that then. Stavrou was merely saying that in order to get the best out of 16 bit digital at the time we should aim for a K reference of K-12 (rms) which is pretty loud and close to 0dB FS.

Notice Ben there is not even one mention of 24 bit recording in that chapter. Your assertion now is completely wrong and you should not be putting up incorrect facts and only creating confusion for new readers. Most of us I am sure agree that what you are saying is incorrect.


Here was my response to his argument back in the Analog VS digital camera thread:


Re:AnalogCameraVsDigitalCamera - March 25, 12 8:51 AM ( #17 )

I may be able to throw some light onto the subject. Ben is quoting a chapter from Mike Stavrou's book 'Mixing with your Mind' from the chapter 9 Digital VS Analogue. Mike uses the analogy of a large skyscraper. He says that because of its enormous height it is hard to capture the whole thing in focus. He is saying the best sound from analog is in the middle of the skyscraper and this is similar to level. Right at the top transients and things will get distorted and right at the bottom the tape hiss starts to mask low level information.

In Digital recording however Mike says the very top of the skyscraper is the cleanest and sharpest part of the picture and it gets progressively blurry as you go down to the bottom. Level wise we know that the maximum number of bits is used when our signal is high and as we go down lower in level we are using less bits to capture important information.

Mike is saying that setting your digital reference level way down at -18 or even -20 db FS as Bob Katz points out may not be the best idea because we are not using all the available bits to capture most of the music or where most of our music lies level wise. Mike also suggests that a better digital ref level might be the Katz -12 db FS instead where more bits are being used. But this requires you to lift your entire gain structure up there and also requires some effort in preventing clipping as we now only have 12 db of headroom. (I actually work at -14 a lot of the time)

But I must also remind Ben that Mike's book was written in 2003 and I really get the impression he is referring to 16 bit recording most of the time and if you are working at 16 bit level then what Mike is saying is probably true. But things have changed a lot since then and we are now in a different situation where 24 bit recording is common place. The skyscraper is different now and much more of it is in focus and the bits down the bottom are now -144 dB away from the very top of that sharp focus image.

So Ben a very recent article in the said Audio Technology magazine has completely debunked this concept of having to record at higher levels in the digital world and with 24 bit recording we are not under that pressure anymore so the -18 dB or Katz -20 dB ref level is perfectly acceptable. You are not loosing any detail now even at -20 because the noise floor is now still -124 dB below that. In the 16 Bit world if you are using a ref level of -20 db with only 96 dB of dynamic range (and we know we really only have -90 dB  in fact)  then the noise floor is only sitting 70 dB below that which could be considered dangerous in terms of very important low level harmonic material. Not so with 24 bit though.



We have just aquired a new mixer at the TAFE where I teach. It is an SSL AWS 948 console and close to $100,000. An SSL guy came all the way out from th UK to teach us how to use it. (I must say it is rather nice!) Anyway it has VU meters on the front. The reference levels that SSL recommend are switchable and are -24, -22, -20 and -18. Of course the -20 agrees perfectly with the K standard. I don't think SSL would recommend using a ref level that is even lower then the K -20 (eg -24) if it was SOOOO wrong.

If you are going to post things in the Techniques forum it is essential that at least the information is as correct as it can be.

The problem is the highest volume adda converters use is around 118db no adda converters use the full dynamic range of 144db.  In fact motu and rme use between 110db and 118db  Presonus have 118db of dynamic range. And lexicon us series are 101db. Pretty close to full range 96db. 

Once again you should not misslead readers. Nowhere in that article does stav mention 16 bit but it doesn't matter if adda converters still do not use the full 144db of theortical dynamic range of 24 bit.  

The way I interpret it is the theory is sound then and now.

I would like to see imperical evidence pg number and the exact quote of 16 bit. 

How can the theory be turned on its head due to bit depth ESP if 24 bit does not mean 24 bit. At the moment it's more like 20 bit. 

I have technical degrees you have musical degrees leave the technical stuff to me and start to worry about the relevance of your course as I hear tafe courses are about to be cut in Victoria and the relevance of Music Industry Technical Production is irrelevant and one of the to be first cut.  You can take that to bank!!

Ben

I accidentally recorded a whole song in 16 bit without realising once. I put lots of compression on all the tracks and mastered it loudish so brought that noisefloor right up. I never even noticed till one day I was looking at the original wav files and noticed they were 16 bit. I listened back to the song hard and only just managed to hear it. I could, but barely.

My point here is there are so many other things that make SO much of a difference. We stress about 24 bit so much, but 99% of people would never even notice if you don't tell them. The difference in noisefloor from 16 bit and 24 bit is audible, but the 16 bit is still actually pretty darn low. You're only going to hear it in a song where only one instrument is playing, not when the whole thing kicks in. That noisefloor, even if it kicks up to say -60dB after mixing and mastering, it's going to be pretty well masked 90% of the time. How many of you actually have a background level and equipment noise level when recording that is LESS than -96dB? Not many.

I do find it interesting doing these tests and seeing what is better etc, but at the end of the day, that's all it is - interesting. You're talking a 0.2% increase in the sound quality of your song. Double your noise isolation when recording and you've already overshot the losses in the digital domain by far. Change mics and that distortion that you get from recording down low (if it is the case) is minuscule in comparison. I personally would rather just quickly set things safe so you don't get overs and get down to recording some great music. Stressing about getting things as high as possible then being really meticulous about your playing so you don't accidentally go over could ruin in the feel of your music. Just turn the gain down, stress less, put out a great performance, and work on the sounds and quality that really matters and makes a reasonable difference.

Very well put Matt I could not agree more. Even though I may go on about these things, they are insignificant in the scheme of things especially when compared to the music itself.

It's a boring argument this, do we record loud or softer and do we do it 16 bit or 24 bit. I have made great recordings at all levels and bit depths. It is good to get things right engineering wise for sure and you don't need any qualification to be able to do it either, just experience.

As Ben says we should leave all the technical details to him. I am happy to concentrate on the music instead. Once you get the engineering stuff right and it is easy in my opinion, you really need to concentrate on the music and the music only. That is the only thing that stays behind in the final recording, the music only. No one is really that interested in how you got there, it is more about what is there. Technical production can be sloppy or precise and also it is finite in some ways. There are limits on the ways we produce audio.

Our TAFE is doing very well. We are spending money and our course is getting bigger and better. Sure some TAFE sectors are going to be effected and I don't agree with the way our current Premier thinks about TAFE education. But even if things there change I can always increase the amount of music I compose at home and get paid for it. I am pushing that up and actually trying to move away from the other a bit.

But you have got to be reasonably good to create music and get paid well for it too. Something that requires a lot of creativity, art, performance and hard work too. The technical work. And maybe even a music degree or two doesn't go astray. You don't need those to make good music, but they do open doors into new territory. Music is unlimited and a constant learning experience and that is where the art is, not so much engineering art. That is boring art in some ways.

People get too worked up and too involved with minute details of engineering while the music is boring. Spend large amounts of time looking right into the music and all the ways you could make it better. Every aspect up to where the microphone is and don't worry about from that point on. The initial creative element could be done fast as some great works of art are created quickly or put it together slowly, improve it later and spend the least amount of time engineering. Good engineers do it fairly quickly. Everyone forgets that magic interaction between good music and good engineering. When the music is great all the engineering easily falls into place. If you are having a hard time engineering any part of your own or anyone's music it means the music is not good or good enough at that point. Re evalute the musical art at that point instead and the engineering will then sound amazing.

If you are spending more time on the software than the music, you have got the wrong program, if you are spending too much time ranting over insignificant engineering details you have got your priorities wrong. At the end of the day it is only your music that is going to effect the listener emotionally and if you want it to, sell you or not.

I think it something to do with age maybe. When I was younger I was really excited and fascinated in the engineering process and all the techncial stuff and I still enjoy it of course but at some point (when you get older or at a certain age) I think you have to just leave it all alone and only think about the music.  New music. Great music. We have to come up with fresh ideas and musical concepts or that is how I feel about it anyway.

I just want to perform my music well enough to get in the songs forum...no picking of nits here...

I personally think this will always be a subjective topic for everyone. My advice is to simply do a few small projects while logging your inputs and then determining at the end which is best for you and the soundcard/style you're dealing with.

I've tried every input setting I could mess with and for me personally, -6dB peak seems to be the best bet for me no matter what bit or sample rate I record at. Even an average of -6dB to where certain things may hit -4dB or even -3dB have been fine for me. But these days I've been really happy with no hotter than a -6dB peak. If a person gets good sound at the hottest point they can record digitally, I say go for it. If you prefer -10dB or less, go for it. Let your ears be the judge. :)

-Danny

First sentence of first paragraph (please correct me if I'm wrong)

I am by no means an expert on digital signal processing but

That says it all.

Ben, record all your tracks at 0dB, mix them together at 0dB, now try to get them out of the DAW WITHOUT turning them down (and, according to your expert creating distortion).
What you are presenting here is the "RESOLUTION" argument; it doesn't work.
It DID NOT work at 16 bits, when sig>noise MIGHT be an issue.

0vu = +4dBu (1.23v rms) = apprx -18dBfs (depends on the the DAW)
So to get your line signal at 0vu (this is analog now) into the DAC so it can go into the DAW at + 18dB.
YOU GET DISTORTION.

Anyone who suggests recording "as close to 0 as possible" is NO AUDIO EXPERT!
RUN don't walk away as quickly as possible.
read this thread at GS Gain Staging Mixing ITB (I took some liberty for the sake of brevity)
Absorb everything Skip Burrows and Paul Frindle said.

Tom

The error in this argument, like many errors involving audio, is that it assumes that humans have infinite resolution to hear quantization noise/distortion/error, and/or are recording through a perfectly noiseless analog chain and/or are recording in an environment with no (desirable) background noise.

In the real world, once something is of such low level that it will never be audible to humans and/or is significantly below the recorded noise floor, the word "better" no longer applies.

So you end up with people making heated arguments that are entirely hypothetical/theoretical and have nothing to do with the real world.

This is made much worse by the fact that humans have been proven to be both quite incapable of determining what they hear vs. what they imagine they hear, and routinely imagine they can hear things that aren't there.

Spot the 'real' approach.




I love that picture.  Thanks Ethan.

Now if you'll excuse me, I have no desire to feed the troll.

It's easy enough to figure out how to do a test on relative input levels, it doesn't even need another discussion.  How much better than I can't hear a difference do you need?

If you have failed to set up a proper test or cheated and find a difference then you have to determine which one is 'best'.

Good luck with that, I'm off to do something more worthwhile like clip my toenails.