Best Recording Levels

[Shifted Music]

Well I for one have noticed increase in bass definition and width utilizing more headroom than less. This is in regards to mixing in the box and for when hitting master faders, busses or whatever.

Purely subjective... no null tests or anything. Just my ears which are easily discounted by anyone wanting to justify how they do things.

Maybe its old school thinking since I have been digital since 1998.

But the important part is - it works for me.

When I have read the breakdowns of why lower levels and more headroom are recommended in regards to voltage and converters and +4 gear interacting with the converters optimum operating levels. It makes sense to me.

[DrFord]

I am remembering this thread and smiling now because OG and I have become good friends... and I have learned much from working with him these past year and a half.

As a related topic to this, I am actually recording drums quieter these days into digital, because it preserves the attack a bit better. I have found in my current setup that if I push drums a bit hard, the mic pres and A to D converter are clipping a bit off the initial transient. Alternatively, I have been pushing vocals a bit into my Apogee AD 16x, with the "soft limit" engaged. By engaging the limiter on purpose, I am finding it to deliver a very nice compression character that has been working well with rap vocals specifically. I end up using less compression once in Pro Tools, but it sounds good to me!

reading back on certain thoughts in this thread, I have definitely grown as an engineer since then, but I still tend to record things hot!

Doc

[PTUser NYC]

This is an old thread.

But it's true that recording at 0dbVU / -18dbFS is the way to go. Here's a bunch of reasons why.


NO INCREASED BENEFIT TO HOTTER LEVELS

When you record a signal, it will already have some noise in it, from the mic, the mic preamp, the electronics it passed through etc. If you can get that noise as low as -75db below the peaks in your signal you're a pretty good engineer. Usually, we all have more noise than that.

When you decide on a recording level for your signal, as long as the noise in your signal is louder than the noise in the recording medium, then you've made the best recording you possibly can.

In 24 bits, most converters are good to a noise floor of about -120dbFS. This is due mostly to electronics noise in the analog input stage of the converter. This means that even if you record your 75db Signal/Noise audiophile signal with peaks as low as -44dbFS (almost off the meter!), you'll still get ALL the resolution in your source file. Your source noise will be at -119dbFS and mask the converters noise below it.

Recording any louder will not increase the signal to noise level, because the ratio of 75db in your source remains, no matter how much louder you record both the peak and the noise together.

There is no noise benefit to recording anywhere on the meter, as long as you get the entire dynamic range of your source.


MORE BITS = LESS NOISE, NOTHING ELSE

People will claim that making a hotter recording "uses more of the bits" and produces a better recording. This is true, and not.

It's true that the higher up the meter you go, the more digital quanta (numerical levels) are available to describe the source. This in turn provides better "vertical accuracy" for digitizing the wave. Sure, I agree. But what exactly IS vertical accuracy and what is the effect of vertical errors?

When a sample is digitized, an instantaneous level is recorded for the voltage in the source at that moment. The level is rounded off to the nearest available digital quanta. That means that there will be some error between the actual level in the analog source, and the digital numerical level that is closest to it.

If you record lower on the meter, you will have greater sized errors in proportion to the signal. True.

But what are the results of vertical errors? They are like an added signal of random vertical fluctuations no greater than half the size of a quanta (or it'd get rounded to another quanta which was closer). And what are random vertical fluctuations? NOISE!

Yup that's right, every time you double the number of quanta (by adding a bit to the word for example) you make the random fluctuation half as large, and that lowers the noise floor 6db! -6db means half as loud.

So a 16 bit system has 96db s/n (16 bits times 6db each), and a 24 bit system has 144db s/n (24 bits times 6db each)

So basically we've just gone through a very fancy way of saying that if you record lower on the meter, the noise floor in the converter is going to be higher compared to your peaks. Yup, we knew that already. And like I said up top, as long as the noise in our source is recorded above that converter noise, we're golden.

So "not getting all the bits" means the same thing as "the noise floor will increase a little, but it will still be waaaaayyy below the noise in my source, so it'll be negligible."

There is NO other concern to not getting full levels other than noise. Anyone who tells you digital audio is a staircase doesn't understand anti-aliasing reconstruction filters. But that's another topic for another post.


IF THERE'S NO BENEFIT TO LEVELS, WHY -18dbFS IN PARTICULAR?

When the powers-that-be standardized digital audio, they knew they had to figure out a levels issue. They had analog tapes with a defined 0dbVU, and of course most recordings peaked over it too.

They couldn't align the top of the digital meter with 0db on the VU meter, or everytime the analog recording popped over, youd get awful flatline distortion on the converter. So they decided to make -18dbFS = 0dbVU.

Let's say we're recording a vocal. It's 1987. We have a trusty U87 plugged into a Neve 1073 mic preamp, and the output of that is going right to a Studer 24 track. We set the levels, and we've got a good level going to the Studer around -3dbVU to 0dbVU with a few little peaks. Everyone is happy. The Neve mic preamp is operating in it's sweet spot, and it sounds great.

Now we take that same output in 2010, and plug it into our converters. We see a level between -21dbFS and -18dbFS with a few peaks up to -15dbFS. What should we do?

I say we're perfect, but the naysayers will tell you to turn up the mic preamp. They want peaks at -5dbFS or higher. So to chase phantom results, they want you to run your Neve 10db hotter! Well, now we WILL have an audible result. The Neve will sound dirtier and more pinched. Sure that can be a great rock and roll sound, but on every track all the time?

Imagine how that'd look on the Studer back in '87 - man you'd better have the trust of the people around you to pull that off. That meter would be PEGGED. yes the tape would be slammed the most, but the preamp wouldn't be so happy either.

So in the pursuit of fictional results, we're ALL supposed to drive our analog devices into distortion? And you think Pro Tools sounds thin because it's digital? Maybe it's because we're recording too hot! Maybe that thin sizzle is the sound of analog mic preamps being pushed too hard, and not digital at all?

Now to be fair, in the early days of digital there were a lot of problems. Good anti-aliasing filters weren't widely available. 16 bits and inferior converters meant that my speech above about noise floor wasn't always true, especially for ADATs and such. There were reasons to record hotter then. But that's not been true for a long long time.


INCREASED HEADROOM

When you record hot, you not only record more distorted sounds, you also eat up headroom for processing. If you're recording a vocal within an inch of Full Scale, there won't be much room for a plug in to do anything. You can't boost 4k 3db if your vocal track peaks at -2dbFS.

I gotta say, the mix bus seems to breathe better too when there's increased headroom. I usually see my mixes peaking around -10db or less. Try it, you'll be surprised at how good Pro Tools can sound.


BETTER INTEGRATION WITH ANALOG PROCESSING

If you record above -18dbFS, and then go out of Pro Tools into the analog world, then that gear is going to want to see a 0dbVU level or it'll distort and not sound so sweet too.


BUT I WANT A LOUD MIX

Of course! We all compete in the world of loudness wars. You can always put a plug in last in the chain on the master fader and raise the whole thing up. I like to use a compressor here anyway.

And yes, that will raise everything, the peaks, the noise floor in your source tracks and the noise in the converter. Again, its the ratio between these things that matters, and it remains the same - the peaks to the noise in the source tracks.

I'm not even saying don't compress. If you're making a modern rock record with a very limited dynamic range, it'll still sound better if you smash all those channel compressors etc around 0dbVU. Smash away for aesthetic's sake, but do it at lower levels so you have control.


CONCLUSIONS

You don't need to record hot. There is literally ZERO benefit.

Things sounds better at lower levels because of better performance from the analog front end.

Things sound better at lower levels because of better performance from digital processing and summing.

You can always bump up the superior result to compete.

We used to have to work at this SO hard in the days of tape. Now, we have low noise tools that literally make it a no-brainer to make clean recordings. We literally CAN'T screw it up unless we record SO low we're off the meters, or too loud.

Why are we all talking about recording too loud then? Its the ONE thing we could do to screw it all up.

[ericlees]

agree...and besides the noise of the recording, always you're on digital domain, there's also the signal to error value, sometimes confused as noise to signal.

[Picture Start]

-18dBFS = 0VU standard was good for tape machines and digital recorders to do transfers. If -18dBFS equals +4dBu then 0dBFS equals +22dBu, an operating level you're not likely to tax circuitry in the analog domain. Your highest peaks on tape calibrated to +6/185 would likely tickle 0dBFS but probably not exceed it.

[audiogeekzine]

I'm happy this thread got bumped.

Engineer Todd Burke recently wrote an excellent article on this subject.
http://www.toddburke.com/tb_pages/To...Structure.html

[Stig Eliassen]

Originally Posted by audiogeekzine

I'm happy this thread got bumped.

Engineer Todd Burke recently wrote an excellent article on this subject.
http://www.toddburke.com/tb_pages/To...Structure.html

Great article. Thanks!

[Stig Eliassen]

A big thank you to PTUser NYC too!

[PTUser NYC]

Originally Posted by PT Lover

A big thank you to PTUser NYC too!

Sure :) this is like religion for me. Pro Tools can sound VERY VERY good, and making a clean recording in Pro Tools is SO much easier than on analog tape. And yet, a lot of people do the one thing they shouldn't and screw it up.

IF you absolutely NEED to record hotter, at least calibrate your converters so that the analog side is in step, and watch for overs on plug ins and buses.

Personally, I think good engineering practice says to keep headroom in summing, even for projects where everything is distorted and smashed.

[mightyduck]

page 370 of the pro tools hd guide [v8]

7 Adjust the output level of your sound source
(instrument, mixer, or preamp). Monitor the
track’s meter levels in Pro Tools to ensure that
levels peak within at least –6 dB to –12 dB on the
input meter without triggering the clipping indicator
on your audio interface.


You have some good info in your post, but I believe you have misstated or neglected a couple of things.

Originally Posted by PTUser NYC

When you decide on a recording level for your signal, as long as the noise in your signal is louder than the noise in the recording medium, then you've made the best recording you possibly can.

That is simply false. In truth, you should try to get your signal as far above the noise as possible [without causing other problems, such as clipping or distortion of the adc]

Originally Posted by PTusernyc

]In 24 bits, most converters are good to a noise floor of about -120dbFS. This is due mostly to electronics noise in the analog input stage of the converter. This means that even if you record your 75db Signal/Noise audiophile signal with peaks as low as -44dbFS (almost off the meter!), you'll still get ALL the resolution in your source file. Your source noise will be at -119dbFS and mask the converters noise below it.

That's not correct. Johnson noise and quantization distortion have different characteristics. Quantization distortion tends to be much more intrusive, and is not necessarily "masked" by analog noise.

Originally Posted by ptusernyc

Recording any louder will not increase the signal to noise level, because the ratio of 75db in your source remains, no matter how much louder you record both the peak and the noise together.

See above.

Originally Posted by ptusernyc

There is no noise benefit to recording anywhere on the meter, as long as you get the entire dynamic range of your source.

Again, simply false. See above.

MORE BITS = LESS NOISE, NOTHING ELSE

Originally Posted by ptusernyc

People will claim that making a hotter recording "uses more of the bits" and produces a better recording. This is true, and not.

Its simply true, in the sense that the accuracy of samples will be better. You only get one chance to do the initial sampling.

Originally Posted by ptusernyc

It's true that the higher up the meter you go, the more digital quanta (numerical levels) are available to describe the source. This in turn provides better "vertical accuracy" for digitizing the wave. Sure, I agree. But what exactly IS vertical accuracy and what is the effect of vertical errors?

When a sample is digitized, an instantaneous level is recorded for the voltage in the source at that moment. The level is rounded off to the nearest available digital quanta. That means that there will be some error between the actual level in the analog source, and the digital numerical level that is closest to it.

If you record lower on the meter, you will have greater sized errors in proportion to the signal. True.

But what are the results of vertical errors? They are like an added signal of random vertical fluctuations no greater than half the size of a quanta (or it'd get rounded to another quanta which was closer). And what are random vertical fluctuations? NOISE!

Yup that's right, every time you double the number of quanta (by adding a bit to the word for example) you make the random fluctuation half as large, and that lowers the noise floor 6db! -6db means half as loud.

So a 16 bit system has 96db s/n (16 bits times 6db each), and a 24 bit system has 144db s/n (24 bits times 6db each)

So basically we've just gone through a very fancy way of saying that if you record lower on the meter, the noise floor in the converter is going to be higher compared to your peaks. Yup, we knew that already. And like I said up top, as long as the noise in our source is recorded above that converter noise, we're golden.

So "not getting all the bits" means the same thing as "the noise floor will increase a little, but it will still be waaaaayyy below the noise in my source, so it'll be negligible."

There is NO other concern to not getting full levels other than noise. Anyone who tells you digital audio is a staircase doesn't understand anti-aliasing reconstruction filters. But that's another topic for another post.

That's incorrect. There is less noise and the accuracy of the samples' amplitude is improved. The ear is extremely sensitive to variations in pressure, notwithstanding where your threshhold of hearing may be.

Also, you seem to be completely ignoring sounds that are synthesized within the daw, thus having no analog noise component.

Originally Posted by ptusernyc

When the powers-that-be standardized digital audio, they knew they had to figure out a levels issue. They had analog tapes with a defined 0dbVU, and of course most recordings peaked over it too.

They couldn't align the top of the digital meter with 0db on the VU meter, or everytime the analog recording popped over, youd get awful flatline distortion on the converter. So they decided to make -18dbFS = 0dbVU.

Let's say we're recording a vocal. It's 1987. We have a trusty U87 plugged into a Neve 1073 mic preamp, and the output of that is going right to a Studer 24 track. We set the levels, and we've got a good level going to the Studer around -3dbVU to 0dbVU with a few little peaks. Everyone is happy. The Neve mic preamp is operating in it's sweet spot, and it sounds great.

Now we take that same output in 2010, and plug it into our converters. We see a level between -21dbFS and -18dbFS with a few peaks up to -15dbFS. What should we do?

I say we're perfect, but the naysayers will tell you to turn up the mic preamp. They want peaks at -5dbFS or higher. So to chase phantom results, they want you to run your Neve 10db hotter! Well, now we WILL have an audible result. The Neve will sound dirtier and more pinched. Sure that can be a great rock and roll sound, but on every track all the time?

Imagine how that'd look on the Studer back in '87 - man you'd better have the trust of the people around you to pull that off. That meter would be PEGGED. yes the tape would be slammed the most, but the preamp wouldn't be so happy either.

It seems to me that your figures are somewhat "doctored" there. Using pro equipment you are just not going to be overdriving stuff in the way you describe.

You seem to be saying that the meters on a daw are to be read the same way as a VU meter. That's not correct. DAW meters are generally peak meters. Peaks of -6dBfs or so may generally give you the -18dBfs RMS reading that you want to correspond to 0VU.

Also, as you noted in a subsequent post, there are trim controls, and the solution is not to record unduly low, but, rather to adjust the trims on your interfaces to optimize both digital level and analog level. My 192 interfaces came from the factory set at -15dBfs = 0 VU. Rental houses will typically calibrate to -14dBfs or -15dBfs for music work. Film work can be a special case on a soundstage, or live field recordings, where you may be faced with very unpredictable loud sounds, and -20dBfs is common there.

So in the pursuit of fictional results, we're ALL supposed to drive our analog devices into distortion? And you think Pro Tools sounds thin because it's digital? Maybe it's because we're recording too hot! Maybe that thin sizzle is the sound of analog mic preamps being pushed too hard, and not digital at all?

No. But your premise is incorrect there.

Originally Posted by ptusernyc

Now to be fair, in the early days of digital there were a lot of problems. Good anti-aliasing filters weren't widely available. 16 bits and inferior converters meant that my speech above about noise floor wasn't always true, especially for ADATs and such. There were reasons to record hotter then. But that's not been true for a long long time.

Nothing's really changed. Its just a couple of bits better. By the way, you don't get 24 bits, maybe 19-21, or so.

Originally Posted by ptjusernyc

When you record hot, you not only record more distorted sounds, you also eat up headroom for processing. If you're recording a vocal within an inch of Full Scale, there won't be much room for a plug in to do anything. You can't boost 4k 3db if your vocal track peaks at -2dbFS.

That is a complete non-issue. Most plug-ins have input trims, or you can simply back the signal down in the 48 bit mixer after recording it at an optimal level [or as optimal as is reasonably practical in the circumstances].

Most of what you are cautioning against is easily overcome by simply knowing how to record and mix. The "bad results" from "recording too hot" that we are always hearing about is simply people doing stupid stuff.

Originally Posted by ptusernyc

I gotta say, the mix bus seems to breathe better too when there's increased headroom. I usually see my mixes peaking around -10db or less. Try it, you'll be surprised at how good Pro Tools can sound.

That's probably in your head. Mixes don't "sound better" at -10dBfs than at -6dBfs, or even hotter. As long as your are not clipping or otherwise distorting it is no problem.

Also, recording with "optimally hot" levels does not mean that you have to mix hot. The problem seems to be for people that are pushing it into clipping because they do not understand gainstaging.

Originally Posted by ptusernyc

If you record above -18dbFS, and then go out of Pro Tools into the analog world, then that gear is going to want to see a 0dbVU level or it'll distort and not sound so sweet too.

As explained above, you seem to be confusing peak meters with VU meters.
There is no good reason to record with peaks at -18dBfs. That's unduly low in most cases.

Originally Posted by ptusernyc

BUT I WANT A LOUD MIX

Of course! We all compete in the world of loudness wars. You can always put a plug in last in the chain on the master fader and raise the whole thing up. I like to use a compressor here anyway.

And yes, that will raise everything, the peaks, the noise floor in your source tracks and the noise in the converter. Again, its the ratio between these things that matters, and it remains the same - the peaks to the noise in the source tracks.

I'm not even saying don't compress. If you're making a modern rock record with a very limited dynamic range, it'll still sound better if you smash all those channel compressors etc around 0dbVU. Smash away for aesthetic's sake, but do it at lower levels so you have control.

Another thing I think you are neglecting to take into consideration is the fact that many mastering engineers are going to limit your mix so that the noise / dither / whatever is even closer to the signal.

Originally Posted by ptusernyc

You can always bump up the superior result to compete.

Well you don't really get a "superior result" by recording at low levels, unless you are comparing it to a mix where someone clipped and distorted a bunch of stuff. But you do not have to clip or distort your mix just because you record at hotter levels. Clipped or distored mixes are generally a result of poor engineering, not robust initial recording levels.


CONCLUSION:

Read the manual [see above p. 370 recommending recording "within at least –6 dB to –12 dB on the
input meter"].