Recording Signals, - 18dbfs??

[Jay Dee]

Originally Posted by James Meeker

When starting projects in PT I put Waves PAZ on all the channels so I get accurate peak/RMS levels.

Sonalksis' Free Gain is good for this, too (and... free).

Sonalksis - Quality Digital Audio Software

[R3altruth]

Originally Posted by Jay Dee

Sonalksis' Free Gain is good for this, too (and... free).

Have you tried the RND inspector... I've heard good things about it

[Jay Dee]

No I haven't... have you got a link?

[narcoman]

Originally Posted by colinmiller

I have always found working at -18dBFS to be problematic. I suppose if you're always the only one working on a project you could just discipline yourself to always make 0 be -18. But when you're working with others it's pretty much impossible. And -18 always means the signals getting slammed into the console. And one of the biggest problems is that Pro Tools does not allow you to make unity indicator in the meters. So simply looking at the levels on the faders doesn't give ou any good idea where unity should be. So you end up guessing and it gets messy.

For me personally I have found that with -14 I can keep the console and outboard running comfortably and be more compatible with projects coming in and out.

-18 or -20 (depending on who you speak to) is a standard you have to get used to in movies though..... everyone works to it (well, nearly everyone ).

[R3altruth]

rnd | inspector xl

[Skye]

RND Inspector XL is $125 for students/teachers and $250 for everyone else. Maybe it used to be free, but it's not anymore.

I'll see if I can track down that free one and get it to work.

[R3altruth]

There is a trial version of it that only does stereo monitoring

[colinmiller]

Originally Posted by narcoman

-18 or -20 (depending on who you speak to) is a standard you have to get used to in movies though..... everyone works to it (well, nearly everyone ).

Only those in the movies. I haven't worked at that level in 10 years. I suppose I was not specific in my assumption that we were talking about music, nor did I stop to consider we may have been discussing post work. So to be clear I am speaking only in terms of music work and wouldn't know the first thing about post.

For the others, yes there are plenty of plugins that can monitor these things, but what is needed is something that is shown on all the faders/meters at once. It would be trivial for Digidesign to add and it surprises me that they have not done so. I can only imagine that the reason is since they are peak meters and not VU, it may not be too practical since you can't really average things. But I still think it would be extremely useful.

[synthoid]

Originally Posted by Dean Roddey

BTW, I should say that my statements above about floating point values assumes the DAW uses the full range available. If they are mapping everything from 0.0 to 1.0, then that's not the case, and the available range is limited to only those bits available in the fractional part of the floating point representation.

No no, this isn't right. any 24-bit fixed point fraction can be represented exactly in 32-bit floating point. However, many *additional* fractions can be represented in 32-bit floating point. (I'm talking about values whose integer part is 0.)

For example, in 24 bit fixed point (1.23) represention, suppose you have a value N = 0.7A6534 (hex, 24 bits). In 32-bit floating point, you can represent N exactly. However, you can also represent N / 256 exactly; it is the same representation as for N, but with an exponent that is lower by 8. You cannot represent N / 256 exactly in 24-bit fixed point.

Floating point has much (much) better resolution for small values than fixed point.

The 'spacing between' adjacent floating point values is not evenly distributed. In other words, it is not true that the difference between one floating point value and the next one is fixed (constant). As floating point numbers get bigger (exponent gets larger), the space between them gets proportionally larger. On top of this, denormal floating values add a lot of very small values to the range. It's actually a bit complicated, but you don't really need to worry about it unless you're trying to write code that uses floating point arithmetic.

-synthoid

[Dean Roddey]

A 32 bit floating point value only has like 23 bits for the decimal portion. So it would seem that you would have to give up some resolution relative to a 32 bit fixed point value. I guess 24 bits if the leading 0 is assumed.

In a 32 bit fixed point system would give you a full 32 bits of resolution for the decimal value (minus 1 I guess since you have to represent 1.), and with no ability to go over 1.0 as you can in a floating point system.

So what am I missing here?

[Skye]

I downloaded Sonalksis FreeG and it works!

I apologize for asking a newbie question, but I'm having a remarkably difficult time finding a straight answer in a Google search: Yes, RMS means Root Mean Squared, but does it mean the average level? So if I am measuring RMS separate from peaks, I want to keep the RMS around -18db and check just in case that the peaks are below 0db?

[Dean Roddey]

BTW, another incorrect thing I said above was that if they used 0.0 to 100.0 that would be better. That was dumb. In a floating point representation, there's only decimal digits. So there's no non-decimal digits, there's just an indicator of where the implied decimal point is at. So basically there are 23 bits and the decimal's position in there is indicated separately.

But anyway, it has only 23 bits to work with. If you used a fixed point representation with integral values and you are also doing a kind of 0 to 1 mapping, I assume you'd have one special value to represent 1 and everything else is just 32 bits used to represent the decimal digits. Or you'd give up one bit to represent the leading 0 or 1, and use 31 digits to represent the fractional part, but still a lot more than 23 bits.

If they aren't doing a 0 to 1 pseudo floating point representation, then maybe my theory wouldn't hold.

With 64 bit floating point, you have 52 bits to play with, which is far better than 32 bit fixed point can represent.

[Dean Roddey]

Oh, wait, duh! I was missing something completely obvious. Because floating point value uses a floating virtual decimal position, as the number gets smaller and smaller, all the leading zero digits are dumped and the exponent value is just adjusted.

So, the value:

0.00000123912390

in floating point is stored as:

123912390

And the exponent is just adjusted to indicate that there are five implied leading zero digits.

Oh, you'd think I'd know these things already, being a software engineer. But it's just something I'd not thought out before. It only has 23 bits to play with, but that's only an issue when the values get LARGE, i.e. you start pushing out decimal digits on the right as the value gets larger and larger. But going the other way, it's just dumping leading zeros which it doesn't have to store. Since all values are from 0.0 to 1.0 (baring temporarily exceeding 1.0 internally), they get maximum precision, because they never push digits off the right side.

If a fixed point system is really fixed point, then it cannot do that, so it can't toss leading zeros because it cannot float the decimal point. So it would have to represent the value as (loosely speaking, not dealing with the actual number of digits available):

0.0000012391

So it has to give up precision digits on the right even when going downwards, and it doesn't allow you to go above 1.0 because there's no way to represent that.

Sometimes the line between genius and stupidity are very thin, and I find myself way over to one side of it.

[R3altruth]

Originally Posted by Skye

I downloaded Sonalksis FreeG and it works!

I apologize for asking a newbie question, but I'm having a remarkably difficult time finding a straight answer in a Google search: Yes, RMS means Root Mean Squared, but does it mean the average level? So if I am measuring RMS separate from peaks, I want to keep the RMS around -18db and check just in case that the peaks are below 0db?

Yes root mean squared means your average level....
Think of it as how loud your music really is....
Just because a signal peaks at say -6... the average or perceived loudness is probably a lot lower.... the peak is often times just a quick transient
Like when you hit a snare... the initial hit is louder than the decay

the difference between the average and the peak is known as (drumroll please) the peak to average ratio or "Crest factor"

The more dynamic music is (or the higher transient peaks) will give you a bigger difference between the average and the peak....

Slight compression at key stages can really control some transients
and if done right.. certain transients can be lowered without much change to the sonic character of the signal....

Crest factor can vary from Genre to genre and even song to song... Seeing how certain songs may be inherently more dynamic than others... most of my songs at the end of mixing, and before being sent to mastering have an RMS at about -20 and peaks at about -6 for a crest average of 14

I track in correctly and use compression tastefully in the mix.. and I find that I don't have to really check levels too much after the tracking... and at the end when i am ready to bounce, my levels pretty much always fall near that range

[James Meeker]

Good explanation R3altruth!

[R3altruth]

Thanks Meek

[synthoid]

Originally Posted by Dean Roddey

It only has 23 bits to play with ...

Actually, a 32-bit single precision value has a full 24 bits of mantissa to play with. They do a wonderful trick with the encoding to accomplish this. When the mantissa is non-zero, it is left-shifted (and the exponent decreased) so that the leading bit of the mantissa is always 1. This leading 1 bit is then omitted from the representation (not stored); only the remaining 23 mantissa bits are represented explicitly. The encoding of zero (and numbers that are too small to be normalized in this way) is treated as a special case.

-synthoid

[Skye]

Originally Posted by R3altruth

Slight compression at key stages can really control some transients
and if done right.. certain transients can be lowered without much change to the sonic character of the signal....

A couple months ago, I tried recording the tambourine I bought on a whim. Sometimes if I listen closely, I can pick out a tambourine in a surprising number of rock songs. Anyway, I recorded it with my Rode NT1-A condenser, and when I put the tambourine in the mix, it sounded way too quiet, but I was clipping all over. I figured out that the tambourine was what was causing the clipping, even though it seemed a lot lower in the mix.

I was using Cakewalk 2002 then ($12 off of Ebay), and it had a compressor, but there was no visual interface or metering, and you couldn't edit your settings later. To test it out, you could audition it with the compression, but it would start from the very beginning of whichever track you were applying it to. When you wanted to keep the compressor, it rendered the original file. Gosh, it was a pain. I'm glad I have Pro Tools and Waves Renaissance Maxx now.

Well, my theory is (using my new-found terms) is that the tambourine had an enormous crest-factor, since the RMS was low, but it kept peaking before everything else. I can picture how recording a tambourine closely into a condenser could cause some extreme transients. I understood before how compressors could even extremes in dynamic range, like bringing out the cymbals and quieting the kick and snare on a combined drum track. I also understood how a compressor could imitate sustain on a guitar or bass. But that was the first time I saw how it could also increase the useable volume of a track.

I apologize if this was too off-topic.

P.S. I'm sure glad we look more into the science than this guy: YouTube - Point of View 1: Mixing and Mastering.

[R3altruth]

Originally Posted by Skye

P.S. I'm sure glad we look more into the science than this guy: YouTube - Point of View 1: Mixing and Mastering.

HAHAHA... that was good times...

[Audiop]

Originally Posted by synthoid

I have never seen a shred of hard evidence that conversion quality degrades as you approach the analog level that corresponds to 0dbFS (without going over, of course). I have a bit of experience with A/D conversion in other domains (not audio), and I can tell you that the manufacturers try very hard to achieve distortion-free operation especially at the high end of the converter's input range. It goes without saying that there will effectively be more distortion at lower input voltages, because noise affects the signal more as you lower the signal level. It would be ridiculous to build a converter whose analog input stages broke down and distorted at the point where the S/N ratio is the best.

I have seen lots of claims of anecdotal evidence that converters break down as you get near the 0dbFS calibration point. But I suspect this is mostly over-active imaginations figuring out (incorrectly) how the converters work inside and drawing (incorrect) conclusions.

-synthoid

It's actually typically true, when you're getting close to 0dBFS converters typically have increased distortion. Wether this has a significant audible effect or not I don't know. All it takes is to loop a signal and analyze it with some appropriate software to see this.

If this increase in distortion is from the AD, DA, the built in analog circuitry or the external I/V and buffer circuitry I don't know either. :-)


/Peter

[Dean Roddey]

But 'close to 0dBFS' is a fairly loose term. How close? If you mean -1dB, then OK, that's probably not much to argue about since no one argues for getting that high. But -6dB? I'd be kind of semi-shocked if a converter can't handle a signal that high.

But I'd also have to kind of argue that, if this also applies to the D/A, then anything you listen to on any system is likely distorted in such a fasion since even if it's a well done and dynamic mix, the peaks in the loud parts are going to be pretty close to 0dB, if not right up against it.

Does anyone claim that no converters can actually play back at approved CD levels without audible distortion? It would seem kind of strange. If it's not audible there, then why would it be likely to be audible in the A/D direction? Is there something inherenty harder about the A/D direction than the D/A direction?

[Audiop]

Originally Posted by FeatheredSerpent

So then in fact, we would never, under any circumstances, need a practical (fully attainable) bit depth of more than 16 bit, unless analogue technology developed in such a way that it offered a much lower noise floor, and even then there would be no point as we can't hear below a certain amplitude threshold anyway.

I need a holiday.

Hi, some of my thoughts before you dissapear for your holliday.. :-)

If we use noisy mic's and/or mic at distance and use tape and so on then I agree we will have less than 90dB or so SNR and the 16bit noise will be dominated by the rest of the noise in the chain.. but, if we use a low noise mic close up and/or produce some electronic music then we can have a noisefloor below 16bit. Not that I feel 16bit media is a big limitation but if the goal is real life dynamics with at least the possibilty of reproducing 120dB SPL noisefree peaks then we need at least 20bit media, right?

So in fact 24 bit does offer higher resolution over the same region of dynamic range than 16 bit, as well as giving us that extra headroom by extending digital noise downwards. Ok.
As for analogue noise and our ears' realtionship to that, yes it must be very important, because we are never in a real-world situation where there is absolute silence, so our ears are always expecting something!

Well one can look at it from both ends of the scale I'm thinking, you can either push down the noisefloor or stay at the same "noisefloor-level" and extend the max SPL. If you look at it from the latter point of view for normal signals the resolution would be the same only that we can have a higher output system without bringin up the noisefloor.

Don't bother, it makes no sense whatsoever. I think the old explanation for why 88.2 was better was that there were less rounding errors introduced with a simple division of 2, which makes perfect sense.
But then I also read (on here recently) that this is no longer an issue with modern audio engines/algos, which may or may not be true, I don't know.

But was there ever such a downsampling as just "/2"? Wouldn't that result in aliasing?

You attain higher resolution the further you move away from 0dBFS.

It must be the other way around no? The more you move away from 0dBFS the closer to the noisefloor you get and the less of the DR will be left to resolve your signal.


/Peter

[Audiop]

Originally Posted by Dean Roddey

But 'close to 0dBFS' is a fairly loose term. How close? If you mean -1dB, then OK, that's probably not much to argue about since no one argues for getting that high. But -6dB? I'd be kind of semi-shocked if a converter can't handle a signal that high.

Well, start at 0.1dBFS or -1dBFS and move down to -20dBFS and you will se decreased distortion all the way down. Typically the overall THD is decreased and also higher order harmonics are decreased more.

But I'd also have to kind of argue that, if this also applies to the D/A, then anything you listen to on any system is likely distorted in such a fasion since even if it's a well done and dynamic mix, the peaks in the loud parts are going to be pretty close to 0dB, if not right up against it.

Yes but we have distortion in our ears and the rest of the rig as well and from what I have seen we are most sensitive around 80dB SPL which would translate to -20 to -40 dBFS or so depending on amps, speakers and other factors.

Does anyone claim that no converters can actually play back at approved CD levels without audible distortion? It would seem kind of strange. If it's not audible there, then why would it be likely to be audible in the A/D direction? Is there something inherenty harder about the A/D direction than the D/A direction?

Don't know!


/Peter

[synthoid]

Originally Posted by Audiop

It's actually typically true, when you're getting close to 0dBFS converters typically have increased distortion. Wether this has a significant audible effect or not I don't know. All it takes is to loop a signal and analyze it with some appropriate software to see this.

Right, this is what I meant when I said that I have seen lots of claims of anecdotal evidence (but not the evidence itself).

One reason to be especially skeptical of this idea is that many (most?) converters can be calibrated, so that 0dbFS corresponds to several entirely different analog voltages (typically at least three different signal levels). Now, suppose for the sake of perpetuating the myth () that the converter distorts when it is near 0dbFS when calibrated for the hottest input signals. This is believable, I guess, because maybe the engineer designing the A/D converter just had a bad day when it can to that last signal level, and went home early. But why would it distort when calibrated to the two lower signal levels? If there were an amplifier in the input path of the A/D converter itself, you might expect to see something like that. It's just hard to come up with a scenario that explains why this should happen.

-synthoid

[Audiop]

There usually is an amplifier stage in the input of an AD and yes, changing the nominal levels affects distortion.

I assume that changing nominall levels is about changing the feedback resistors around opamps (speaking about typical converters).


/Peter

[James Meeker]

Originally Posted by Skye

P.S. I'm sure glad we look more into the science than this guy: YouTube - Point of View 1: Mixing and Mastering.

Worst. Instruction. Ever. Aight.

[R3altruth]

it was entertaining though....

My favorite youtube comment for that video is:

"at 1st i thought you was black"