Good ways to make your songs approximately the same volume?

How to do this, that and the other. Share, learn, teach. How did X do that? How can I sound like Y?
RELATED
PRODUCTS

Post

I think we're saying, in different techno speak, that we use metering and/or some familiar process to give finalized songs a similar level. No metering or ears are perfect, and who wants them to be? In the olden days an "album" was produced where the levels may fall and rise a bit depending on the artistic need. Nowadays we can't be bothered with listening to an entire album and it is assumed that there is some magic to guess how loud to make the volume in an imaginary 'playlist'. That is the turf of the volume wars and not what the OP asked. So, back to the original poster, is your song collection meant to be at the same volume? That's a whole other assessment outside of the scope of metering, ears and emotion still apply though... :)

Post

aciddose wrote:Actually you can't simplify it. Of course the only real difference is peak vs. average, but RMS meters are an approximation part-way between the two and are far more common. It isn't possible to meter the average level until you have the complete recording available, and it is only possible to do so accurately with a digital recording.
I'd still like to think different. Because if you use a VU meter in real time with 300ms integration time, you do not measure the "maximum" (or even initial) peak, but the "average" peak of the applied signal.

By your saying, VU's and RMS meters - or in simple form "average meters" - are not usable while realtime recording and measurement. Which makes me wonder why they were used then in the 60ies and 70ies.

RMS meters these days are maybe misconcepted as "average meters" (VU's as needle meters, RMS meters as "bargraphs"), and if we go by electric circuit measurement - then yes I agree... a correct measurement can only be done "after" the recording.

So for sake of simplicity: let's just talk about VU and Digital meters while recording (tracking) and mixing). And "RMS" meters for loudness measurements (mean score) after everything was recorded.


aciddose wrote:If you spent some time outside your musical bubble you probably would have. There is a whole world outside that bubble.
Care to give some examples?

Other than these:
http://en.wikipedia.org/wiki/Weighting_filter

I only further know of the k-weighting that is used in the ITU-R BS.1770 specs that is in pretty much constant use. Filters like ITU-R 468 (which is now used as standard for the Fletcher-Munson courve) and ISO 226 are not as common and day-by-day appications.

Also see:
http://en.wikipedia.org/wiki/Equal-loudness_contour
http://en.wikipedia.org/wiki/Fletcher%E ... son_curves


aciddose wrote:Actually it isn't, because the true average level is radically different. Manually windowed levels are radically different. You're missing out on a whole class of measurements.
Not if we talk about pretty much standardised values (5ms, 10ms, 300ms, 400ms, 600ms, 3s). Else it's indeed a clusterfunk all over.


aciddose wrote:With the VU stuff you're talking about this makes the measurement subjective. Any subjective measurement is equally as useless as your ears. You may find it useful, but it is honestly only assisting your subjective interpretation of levels you could also have done using your ears.
I kind of don't know what you're now aiming at?

Do you aim at the % values on a VU (which are indeed not really accurate) or the combination of using a digital meter and VU with 300ms integration time/RMS meter with 300ms timeframe (post recording)?

Of course a VU is somewhat "subjective" for peak measurement, hence the % values on the meter, and the development of actual digital meters!

From the VU Meter wiki entry:
The meter was designed not to measure the signal, but to let users aim the signal level to a target level of 0 VU (sometimes labelled 100%), so it is not important that the device is non-linear and imprecise for low levels.

...

The VU-meter (intentionally) "slows" measurement, averaging out peaks and troughs of short duration, and reflects more the perceived loudness of the material than the more modern and initially more expensive PPM meters. For this reason many audio practitioners prefer it to its alternatives, though the meter indication does not reflect some of the key features of the signal, most notably its peak level, which in many cases, must not pass a defined limit.

...

Since a VU meter does not reflect the signal peaks for complex wave forms and for varying program material, it might be easy to overload a recording device or broadcast system. To protect against this, a lead or margin of safety may be inserted in the meter circuit. This consists, in effect, in an amplifier making the VU meter commonly 6 or 8 dB more sensitive.
The up to 8dB "sensitive headroom" is actually what was used as max peak with the IEC 60268-10 Type II meters (10ms integration time), since they didn't show the exact peak values either.

So they also only show an "average signal strength" with delayed measurement (5ms to 10ms) window rather than the exact one with a digital meter.


aciddose wrote:
Filtered meters use these filters to give a more "true" representation of a signal.
This is honestly just a joke. Filtered meters do not give you a "true" representation of anything. They are just windowed measurements, exactly like windowing in the time domain. They allow you to pick out parts of the "true" signal, more importantly they allow you to ignore the parts you are not interested in.
A quote from the weighting filter wikipedia entry regarding the a-weighting filter:
In the measurement of loudness, for example, an A-weighting filter is commonly used to emphasize frequencies around 3–6 kHz where the human ear is most sensitive, while attenuating very high and very low frequencies to which the ear is insensitive. The aim is to ensure that measured loudness corresponds well with subjectively perceived loudness.
The same can be applied for the k-weighting filter (which utilizes a lowcut, and a highshelf rather than peak EQ and bandpass).

I said "more true", since loudness measurement tools are "perceived" and therefore subjective to begin with. With plain mathematics (unweighted metering tools) bass intensive material responds stronger on the meter than mid or high frequency content. This gives you a wrong impression of actual loudness of the signal (the sum!). And is the very same reason why engineers were all like "funk this - let's press things to it's limits".


aciddose wrote:What I specified were the subjective choices I use for levels of individual elements which has nothing to do with gain-staging. You can go ahead and imagine it does, but that does not factor into the choices I make.
Then that is your way of working I guess. I'm used to analog mixing consoles, and here the rules are a tad different. At least to my understanding.

Agreed, in reality you can not really set a +4dBu = -18dBFS digital value as reference level (I actually measured between -12dBFS to -14dBFS at +4dBu with an appropriate test signal generator). As can be read with a huge red warning sign on this particular page:
http://www.sengpielaudio.com/calculator-db-volt.htm

But it is somewhat "common sense" now. Whether or not this is good or bad, or introduces more problems than it's worth, lies on a whole different ballpark.

Personally I'm fine with -18dbFS = 0VU and usually work from there (bass intensive material not exceeding 0VU, transient heavy not exceeding -9dBFS digital peak, summing bus not exceeding +2VU/-16dBFS and/or -3dBFS digital peak)



Then again, this is still about "loudness measurement/adjustment", no?
And for this, I still say the ITU-R BS.1770 / EBU R-128 specs are currently the best and most accurate tool you can use for that purpose.
[ Mix Challenge ] | [ Studio Page / Twitter ] | [ KVRmarks (see: metering tools) ]

Post

I'm also used to and regularly using an analog console. I follow exactly the same rules there.

In fact there is absolutely no difference between the two when I'm using them. Signals are always normalized at input to make further measurements and settings accurate. There are some issues related to sends/inserts that you wouldn't deal with in the digital realm, but still I use exactly the same methods. While noise issues aren't as important digitally, there are a whole slew of other factors at play which are in most cases far more important to me anyway.

Regarding "loudness" subjectively, this is where weighting filters make sense in an attempt to measure "perceived" loudness of a complete signal. The problem though is that the circuit our ears and brain uses is far more complex than the approximations we use for audio processing. The fletcher-munson curves are the basis and inspiration for all other psycho-acoustic weightings. An important thing to note about these weightings is that the accuracy is dependent upon the content of the signal being measured. A fletcher-munson curve will be most accurate for solo tones and in many cases also for chords and other sets of tones. For more harmonic rich material other rougher approximations like a K-weighting will be more accurate. For pulses or noise various noise-weighting curves are more accurate.

The critical thing is that these are all subjective measurements which are focused on your ears, while peak and average measurements are objective numbers you can actually work with strictly to get exact results.

RMS and other windowed/weighted measurements are inherently subjective and at best an approximation to average measurements. You mentioned that windowed average measures were used in the 60s, sure. These were not true average though and rather often like RMS were exponentially windowed measurements.

A true average is windowed with a rectangular window, (A + B + C + D + ...) / elements.

RMS instead is an exponentially windowed "moving average", (A*x + B*x^2 + C*x^3 + ...), which is a significant difference.

VU meters are similar.

A true average measure would involve integration of the abs() of the signal between the window edges and scaling of the output by the reciprocal of the window length. That is possible electronically and I know it was done, but not in the studios of the 1960s.
Free plug-ins for Windows, MacOS and Linux. Xhip Synthesizer v8.0 and Xhip Effects Bundle v6.7.
The coder's credo: We believe our work is neither clever nor difficult; it is done because we thought it would be easy.
Work less; get more done.

Post

Do we really need a TRUE loudness measuring? And how can something which is "average" be true? Isn't an approximation enough?

I think there are more important elements in mixing than achieving the PERFECT loudness...

(Don't forget that music is about emotions! If there is no groove, there won't be any even with perfect loudness!)

Post

We don't need "true" measurements, what we need are accurate, objective measurements.

I'm just arguing that peak and RMS are useful on their own, but average (rectangular window) is often ignored in audio processing and I find that extremely useful.

I also mentioned that I rarely (if ever) see the average vs. peak measurements referenced. It is common to see people recommending "look at the average of the peak and RMS and the difference between them" but that isn't the same.

I find it is a very useful thing to take a "window", like a single bar or a whole track without fades, or even with fades and take the average level of the entire thing. That is where the -18db level makes sense. That represents the density and average peak level of the whole section. An RMS measurement won't provide the same result.

RMS however will be useful too, just for different purposes. There is a lot to it all.

It is all useful and I regularly use all these methods when I'm mixing. Just that when I was talking about the levels I'm using, those measurements only make sense using specific measures. If you take the relative peak levels I gave for different instruments and try to use RMS it won't make any sense at all. If you try to use those absolute levels it's extremely unlikely that it'll work because they are not absolute, they are relative. So on and so forth, it is necessary to take that stuff all into account for anything I said to actually make sense or remain objective/definitive. Otherwise may as well just ignore everything I said since it'll all become subjective anyway.
Free plug-ins for Windows, MacOS and Linux. Xhip Synthesizer v8.0 and Xhip Effects Bundle v6.7.
The coder's credo: We believe our work is neither clever nor difficult; it is done because we thought it would be easy.
Work less; get more done.

Post

Oh yes, sound engineer vs programmer! :smack: :party:

Post

Ah... now I'm getting where you aim at...
The definition of the measured signal over a course of time equals "average signal strength", measured (for example) with an RMS (bargraph) or VU (needle) meter.

Actually, this is what the Short Term and Integrated Meter does in the EBU R-128/ITU-R BS.1770 meter. The Short Term meter gives you an readout over the timeframe of 3seconds and how strong the signal fluctuates within that time. The Integrated meter gives you an analysis of the whole stream and is therefore an "average" readout where fluctuations of +/-1LU over a course of (say) 5 minutes might be ideal on the long run. Not ideal for an individual track or 30s jingle, but the longer the stream, the better the readouts.


What we are talking about with "average signal strength" is however is a very basic meant concept with unweighted filters (VU Needle Meters, RMS bargraphs), that people use constantly as definition. Here, the timeframe is the integration time. The slower the needle/bargraph responds, the more "average" the signal readout.

Just like a compressor, where people say "medium to average attack", so we just talk about the "speed" if the needle/bargraph ballistics. A "true average" measurement is therefore over a certain given timeframe (like: 3s, 1 minute, 1 bar, 8 bars, a whole song, 15minutes, etc). Unless I still understand you wrong.


And on that behalf, it's understandable why you can read in tutorials, that "the average loudness of a signal should not exceed...", especially considering a certain timeframe. In case of the old K-System v1, this was usually the section where forte to forte fortissimo parts happened.


Soooooo....

That leaves the thing of plain unweighted needle/bargraph (again, modern day VU / RMS meters in hosts and the likes) and their accuracy. To me, while tracking and mixing, they are as accurate as I can get for this particular job. Why? Because this is what was built into outboard gear as well (some use VU's by the ANSI C16.5-1942/British Standard BS 6840/IEC 60268-17 specs with a reference level, some use QPPM bargraphs with 5ms). So I continue using the combination of Digital Meter and VU to judge the signal strength if it's staying within the boundaries of the used gear.


For declaring the actual loudness of a full mixdown however prior to mastering, while mastering and post analysis as well, I tend to use the ITU-R BS.1770/EBU R-128 meter specs. Due to the fact that I get readouts for the Dynamic Range of the stream, the Short Term Loudness (an "average overview" on the histogram, and maximum on the bargraph), the Integrated loudness (complete time of the production, how much does it fluctuate over a certain timespan) and the maximum digital peak.



I think this is more "objective" and "definite" than using plain digital meters and VU's with max peak hold indicators alone. Though I never saw it "wrong" how I worked over the course of the last decade.


In the end, it's down to definition.
Just like with "what is meant with 'warmth' and 'saturation' ". Definitely interesting, yet confusing thinking. We are both right then.
[ Mix Challenge ] | [ Studio Page / Twitter ] | [ KVRmarks (see: metering tools) ]

Post

Compyfox wrote: ...A "true average" measurement is therefore over a certain given timeframe (like: 3s, 1 minute, 1 bar, 8 bars, a whole song, 15minutes, etc)...
Problem and solution in one sentence - enigma.
But what is the relative environment? micro, macro, bar, song, jazz style, metal style, single-artist album, multi-artist playlist, multi-playlists during drive thru the country in a car...
VU, PPM and Ear

Post

Yes of course I just don't like using the term "average" to refer to RMS or other windowed measurements. They're approximating average but not "true average". I guess I can settle for using the term "true average" when we're talking about measuring with a rectangular window although it seems a little redundant to me.

RMS meters, VUs and so on are of course useful tools and they are best suited to specific tasks. I think we with years of experience take this a bit for granted, but we have developed an intuitive sense for these things. Beginners reading our posts probably don't get all those things we're taking for granted. So it's very difficult to try to provide a concise explanation of a technique without accidentally leaving out all these important details.
kylen wrote:But what is the relative environment? micro, macro, bar, song, jazz style, metal style, single-artist album, multi-artist playlist, multi-playlists during drive thru the country in a car...
VU, PPM and Ear
That is another thing you start to take for granted after a long enough time. It does become more natural though, there is just a lot to learn to take into account. The average metering is just another approximation although it is a lot more definitive than a VU, you still need to select the area you are going to measure subjectively.

There are a lot of other factors at play like the playing style, type of instruments and ambiance/reverberation. I mentioned certain "high average level" sounds like a PWM bass. These contribute a bit to the average measurement but this is actually not a negative thing. A track made up of mostly square signal will show higher RMS/average depending upon its density. You will then be able to see how much dynamic range is used in the track by comparing the "true average" level to the peak level.

By adjusting the envelopes and dynamics in combination with the peak levels to maintain a certain ratio of average to peak you'll achieve nearly equal loudness to an orchestral harp recording showing the same measurements. That was the goal, right?

So although you do need to take some additional things into account, knowing that you have these metering tools available and learning to put them to use at the right time will certainly help to achieve the goal.
Free plug-ins for Windows, MacOS and Linux. Xhip Synthesizer v8.0 and Xhip Effects Bundle v6.7.
The coder's credo: We believe our work is neither clever nor difficult; it is done because we thought it would be easy.
Work less; get more done.

Post

As far as loudness goes (which is not what the OP is asking about) I prefer a jazz club or a small rock club loudness for rock/pop music. Not a sound that is 8 feet away from a 130dbSPL rock bands sound system. Now what metering exclusively indicates that? I believe it is both metering and ears (judgement). But balancing the levels of a 20 song package, regardless of its contents, indicates the need for: judgement, monitoring, metering and testing on various systems. Same as it ever was...

Post

You're talking about absolute then switching to dynamics though.

The atmosphere in a jazz club or whatever is based upon both the absolute amplitude as well as the content being played. Here a big difference is going to be where you set the volume control for the system.

Likewise with "8 feet away from a rock band sound system".

What the thread is about is adjusting the final product (the master) so that the same relative intensity / perceived loudness is going to be apparent to the listener on the same system, say a pair of ipod headphones.

I went a little further and mentioned that I follow some rules from both the mixing/producing field and the mastering field and I iterate between them to get the result I want. Compyfox mentioned this also, as well as the youtube video which was linked. It is good to apply some mastering to your mix that you can switch on/off while working to get an impression of where the master might be. If you can't even get close to what you want with the compressor turned on before you pass it along to be mastered, the mastering engineer is going to wish you dead. :hihi:

I also keep ranting about what is in my opinion an often overlooked option during the production and mastering phase - "true average" levels in both peak and RMS and a comparison to peak and RMS-peak, as well as running averages with the usual VU meters and relatives.

Balancing the levels in a 20 song package doesn't require testing on various systems, that is a whole other issue and I think you're just distracting from the core issue here.

Sure, being aware of the conditions the product might be received is important, but in terms of producing a quality mix and master during production this shouldn't be a core concern.

Don't we have enough distractions without crying over what our master might sound like played back in an arena vs. a dark alleyway on a boombox?
Free plug-ins for Windows, MacOS and Linux. Xhip Synthesizer v8.0 and Xhip Effects Bundle v6.7.
The coder's credo: We believe our work is neither clever nor difficult; it is done because we thought it would be easy.
Work less; get more done.

Post

aciddose wrote:We don't need "true" measurements, what we need are accurate, objective measurements.
Without them, it would be easy for someone to release
music that was too quiet, or too loud, or to harsh,
and be a total nuisance in someones randomized playlist.

These days, music with dynamics, should all be shuffled off
to the end of the playlist, so the volume could be turned up
just once, after all the compressed stuff was finished. :phones:

Post

aciddose wrote:I also keep ranting about what is in my opinion an often overlooked option during the production and mastering phase - "true average" levels in both peak and RMS and a comparison to peak and RMS-peak, as well as running averages with the usual VU meters and relatives.
Is there a plugin that calculates that?

And then, is "true average" really the best? Isn't it possible to apply different measurement techniques to different parts of the song? Depending on the content (more transients, less transients, square waves or saw waves etc.)?

Post

I agree with Tricky-Loops here, this is exactly(!) what the EBU R-128 meter is all about. And why the standard (ITU-R BS.1770 / EBU R-128) was created in the first place.


Else, you're really drifting away with all this, aciddose. We went from types of meters and what they do to a highly technical debate about "time frame measurement" to now even "declaring" loudness by genre (which the DR-Meter tried to do, but ultimately failed), and measuring at different distance (where we now talk about an SPL meter, measuring speakers at given distance).

Either you really write a book about all this, or we simply drop it. Why? Because it's getting darn confused. Actually, with all your post you make me feel like as if I'm using meters wrong. And honestly, I've never read anything you wrote in any book I read about this topic.

YMMV.
[ Mix Challenge ] | [ Studio Page / Twitter ] | [ KVRmarks (see: metering tools) ]

Post

Tricky-Loops wrote:
aciddose wrote:I also keep ranting about what is in my opinion an often overlooked option during the production and mastering phase - "true average" levels in both peak and RMS and a comparison to peak and RMS-peak, as well as running averages with the usual VU meters and relatives.
Is there a plugin that calculates that?

And then, is "true average" really the best? Isn't it possible to apply different measurement techniques to different parts of the song? Depending on the content (more transients, less transients, square waves or saw waves etc.)?
Plugin I'm not aware of although you could create one easily enough. A lot of editors have the ability to measure both peak and rms (as in root-mean-squared, not as in a moving average rms-meter) "true average" levels of a complete clip.

For example, the free editor wavosaur. Load a clip, select a portion of the wave or the whole clip and go to tools->statistics. It'll show "RMS Power" which is averaged in some way over the whole clip.

Unfortunately I'm unsure if they take the "true average" or not because it both doesn't say and I haven't seen the code, but it appears at a glance to take the average of the square of the whole clip, then the square root of that = "true average" RMS. It doesn't have any envelope / windowing settings so I assume that is how they do it, easiest way anyway.

It doesn't offer "true average" peak which would be the sum of the absolute of the entire clip. So a full-scale sine will show -3db rather than -4db, but that isn't too much of a problem anyway.

No it isn't "the best", it is just an extremely definitive number you can use to quantify the loudness of a complete recording. It is certainly best in that respect, as far as "most definitive" is concerned. No guesswork, just a single concrete number. Likewise with the total peak or peak RMS (RMS in terms of moving average, the peak reached by such a meter) are definitive results although you do need to choose the parameters for the RMS meter.
Compyfox wrote:I agree with Tricky-Loops here, this is exactly(!) what the EBU R-128 meter is all about. And why the standard (ITU-R BS.1770 / EBU R-128) was created in the first place.

Else, you're really drifting away with all this, aciddose. We went from types of meters and what they do to a highly technical debate about "time frame measurement" to now even "declaring" loudness by genre (which the DR-Meter tried to do, but ultimately failed), and measuring at different distance (where we now talk about an SPL meter, measuring speakers at given distance).
Of course the standards take some of this stuff into account. I just wanted to explain it in a more simple format.

Not sure about "loudness by genre" or anything like that, to my knowledge I never leaned that way :hihi:

I don't think it is really so technical. It can be, but it is easier in my opinion to just take it at face value. Two polar opposites: peak vs. average, and the middle an approximation between them is RMS or VU meters. Likewise we have for frequency: flat vs. complex curves which approximate human perceived loudness, with simple selective filters (lowpass, highpass, bandpass) between them. K-weighting would actually fall more toward the middle rather than the edge.

Again, the simple explanation is that these different types of measurements all have their own purpose and there is no "best" measurement. By taking into account the range of tools available you can come up with a process which reliably will allow you to produce tracks with "approximately the same volume" without merely leaving it to your ears or subjective interpretation.
Free plug-ins for Windows, MacOS and Linux. Xhip Synthesizer v8.0 and Xhip Effects Bundle v6.7.
The coder's credo: We believe our work is neither clever nor difficult; it is done because we thought it would be easy.
Work less; get more done.

Post Reply

Return to “Production Techniques”