Device for Equal Loudness/Fletcher Munson Curve? Do Any Speakers Adapt to This?

[ernestcarl]

nver heard of the "dialogs quiet, explosions loud" problem?
but yes, it is not all movies...so it is caused by bad mixing, not the standard per se.

also, in general the soundeffects on some movies are just stupid, I remember recently meassuring a movie scene of a closing trash can that had 20 and 30Hz content. I also remember "The Invisible Man" as very bad; for example a guy running into a parked car sounding like a bomb

I do see what you mean. I've not encountered it often. I can recall one documentary show/film downmixed had that very obvious problem -- but that was many months ago. At the time this was more prominent, I may have been using quite bandwidth-limited speakers to have noticed.

 

[mhardy6647]

Here I solved this issue by making my own downmix in a software matrix mixer

I read this -- twice -- as dominatrix.

I have gotta start gettin' out more again...

 

[dasdoing]

I read this -- twice -- as dominatrix.

I have gotta start gettin' out more again...

lol....here in Brazil we say "Freud explains this"

 

[playmusic]

I have browsed through this thread and I think that my following question has not been discussed, yet.

The Fletcher-Munson equal-loudness contours are about sine wave signals for specific frequencies.
As I understand it, the OP asked about DSP of music.

There is a difference between pure sine waves and music.

Consider a soundtrack that is played back with an RMS of 80 dB SPL.
Choose a frequency, say, 50 Hz.
It is absolutely unclear how strong the 50 Hz component is in the soundtrack.
So, it is unclear how strong a DSP correction should be if the soundtrack is played back at another SPL.

So, what is the model for applying F-M corrections to music?

I am looking forward to your feedback!

 

[Aerith Gainsborough]

Consider a soundtrack that is played back with an RMS of 80 dB SPL.
Choose a frequency, say, 50 Hz.
It is absolutely unclear how strong the 50 Hz component is in the soundtrack.

As far as I understand it, that is completely irrelevant.

You lower the volume of the entire signal by 5% and your ears' non linearity will make it appear as if you lowered the bass part by 10%.
The DSP curve tries to account for that by only lowering the bass portion of the signal by 2.5%, so it would appear that you indeed lowered it by 5%, which was your intention. (Numbers are arbitrary, in order to demonstrate)

If done correctly, the spectral balance past your ear will not change, despite your measurement gear saying it did change before your ear.

So far, I have found JRivers loundness compensation to be "hit and miss". On some tracks it sounds good, on most tracks it annoys me. That's probably because I don't use a sub, so I don't get the benefit of more sub bass and only get the bass >40ish Hz amplified.

Also, calibration is a wee bit awkward with my current implementation, because I lack digital headroom.

PS: technically it is not "absolutely unclear how strong the 50 Hz component is in the soundtrack". You can always perform a spectral analysis via Fourier transformation to find out.

 

[dasdoing]

Consider a soundtrack that is played back with an RMS of 80 dB SPL.
Choose a frequency, say, 50 Hz.
It is absolutely unclear how strong the 50 Hz component is in the soundtrack.

it doesn't matter,
let's say it is barely audible at 80dB in the context,
it will be inaudible at 70dB
THAT is the point

 

[playmusic]

You lower the volume of the entire signal by 5% and your ears' non linearity will make it appear as if you lowered the bass part by 10%.
The DSP curve tries to account for that by only lowering the bass portion of the signal by 2.5%, so it would appear that you indeed lowered it by 5%, which was your intention. (Numbers are arbitrary, in order to demonstrate)

Thank you for your reply.

Your line of reasoning would be the answer if in the diagram in the first post of this thread for any vertical line (corresponding to a frequency) the intersection with the shown contours would be equidistant points.
But this is not the case.

The (arbitrary) numbers in your example only apply to a certain dB SPL level for a specific frequency.
The level changes, the numbers change. As an example for the red contours: For 20 Hz, the difference between the threshold and 20 phon curve is 18 dB, whereas it is 8 dB between the 40 and 60 phon curves.

PS: technically it is not "absolutely unclear how strong the 50 Hz component is in the soundtrack". You can always perform a spectral analysis via Fourier transformation to find out.

Indeed, a spectral analysis is necessary to find the level of the 50 Hz component.
When all the information is "A soundtrack is played back with 80 dB SPL" then this would not be sufficient to determine the 50 Hz component. The component would be unclear/unknown.
So, the required DSP depends on the specific soundtrack and on the specific playback level.
From what I have read in this thread, the idea of the forum members posting in this thread about the DSP seems to be that it should independent of the specific soundtrack.

That is why I am still wondering: What is the model for applying the contour corrections to a soundtrack?

 

[dasdoing]

ok, I get your point now,

first, it was never sugested that it creates perfect compensation.....for this the algorythm would have to be made for the specific listener ear anyways.

second, while not exactly true, the "spacing between the levels" are kind of linear.
the following is a graph of the ISO 226:2003 curves, relative to a reference level of 90dB (the flat uper border).
if you look at your 50Hz and the reference frequency of 1000Hz > let's say we have the 50Hz at 90dB and the 1000Hz at 80dB. now we attenuate 10dB via the algorythm. you get -6ish for 50Hz and -10dB for 1000Hz.
now if you do the same for a starting point with 50Hz at 84ish and 1000Hz at 90dB. now we attenuate 10dB via the algorythm. you get about 6ish for 50Hz again and -10dB for 1000Hz.

it obviously will never be perfect.....but it will be much better than hearing only half the music.


but hear for yourself: listen to Rebecca Pidgeon's rendition of Spanish Harlem -40dB below you reference level,
and then compare to this -40dB loudness compensated version (at your reference level): https://drive.google.com/file/d/11LjGW8E4Uqx4X17oKYvToJQrGTbzy-w9/view?usp=sharing

 

[playmusic]

it obviously will never be perfect.....but it will be much better than hearing only half the music.

Thank you very much, @dasdoing for the explanation!
I understand now the approach.

I listened to the Spanish Harlem version at Rebecca's youtube account.
The original has an RMS of -18.5 dB.
Your linked uploaded compensated version has an RMS of -48.3 dB.
To match RMS, I listened to the original at -30 dB.
I agree that the compensated version sounds more similar to the original.
(Listening to the original at -40 dB is as expected the least pleasing version for me.)

My remaining question is about the default reference level dB SPL for music - to which no compensation needs to be applied.

There are some posts on this forum about reference levels, but I still wonder if there is an agreed industry standard.
Under the assumption that a standard of r dB SPL exists, I would also have the follow-up question how it is to be interpreted.
Most music is mixed for speakers.
If it is stereo then sounds from two speakers reach each ear.
So would the standard be r dB SPL per speaker?
Or is it expressed as r'=r+3 dB SPL at each ear for the sum of incoherent signals?
Would it be r for speakers and r+3 for headphones?

(I would have thought that left and right channel were at least partially coherent, but a quick check with Spanish Harlem showed that they are not.
So, the sum is not x+6 dB SPL or something in between x+3 and x+6 dB SPL. It is exactly x+3 dB SPL).

 

[Aerith Gainsborough]

IIRC our ear is most linear at 83dB. So I would "calibrate" the system to that value. Or to be more precise: I would set the system so that the compensation does nothing at that value.

HOW to calibrate exactly is a topic of discussion. Some say you need to calibrate using a -20dB (FS) pink noise test signal in order to simulate actual music. Others say that it is too extreme. You should find some discussion in this very thread about the topic.

I don't think they take # of speakers into account. They simply measure the volume at the listening position.
Most music has very different left/right channels, depending on the placement of the instruments / mics.

 

[dasdoing]

there is no real reference level for audio production: https://www.audiosciencereview.com/...sure-level-flowchart.11069/page-2#post-417437

what engeniers do is to reference to other songs....so in a specific genre the level where sound is balanced should be close from one song to another.
change the genre and the level will probably not work anymore.

personaly I used a pink noise reading -14 LUFS (since this is kind of a standard now because of streaming services) and calibrate this to 83dB.
but this was only a starting point. I tweaked it a little when I felt my music was not balanced at the 0dB level.
what I also do is using a LUFS meter for non streaming services sources to then adjust the level to -14 LUFS

 

[dasdoing]

unfortunaly the most cientific way to set reference levels is by ear, because we don't know the real one. set it where your music sounds the most balanced. it would also be best to have specific ones for every genre

 

[playmusic]

I used a pink noise reading -14 LUFS (since this is kind of a standard now because of streaming services) and calibrate this to 83dB.

I looked for a definition of LUFS but gave up. Internet sources state that it is connected to RMS, but different. Some state that Fletcher-Munson correction is applied before calculation of LUFS. Others state that LUFS is equal to RMS but only for levels below -7 (i.e. -10 or -20).

Could you please direct me to a definition of LUFS?
Or at least explain what -14 LUFS pink noise mean? Is it pink noise with an RMS of -14 dBFS? (And what version of RMS you use: Is RMS(FS sine) equal to 0 dBFS or -3 dBFS?)
Thank you!

 

[dasdoing]

I looked for a definition of LUFS but gave up. Internet sources state that it is connected to RMS, but different. Some state that Fletcher-Munson correction is applied before calculation of LUFS. Others state that LUFS is equal to RMS but only for levels below -7 (i.e. -10 or -20).

Could you please direct me to a definition of LUFS?
Or at least explain what -14 LUFS pink noise mean? Is it pink noise with an RMS of -14 dBFS? (And what version of RMS you use: Is RMS(FS sine) equal to 0 dBFS or -3 dBFS?)
Thank you!

LUFS is a scale to measure audio programme loudness. you can think of it as a better version of the antique RMS.
the standard is described here https://www.itu.int/dms_pubrec/itu-r/rec/bs/R-REC-BS.1770-4-201510-I!!PDF-E.pdf

-14 Lufs is a level used by Spotify (and other streaming services). if you have normalization turned on all music will be leveled to -14 LUFS (integrated).
I have send my pink noise through a LUFS meter and leveled it until it read -14 LUFS (*).
this is a method I came up with, but I never saw anybody else use it....so take with a grain of salt lol


(*) because I figured to have constant loudness the best way is to adopt this level to other sources, too.....I will ajust other sources with the LUFS meter during playback

 

[playmusic]

@Aerith Gainsborough Thank you, it is indeed about the SPL at the ear position.
I also looked now at source [1] in the first post of the above reference loudness thread.
Section 4.1.4 gives the summation formula for incoherent sounds of equal level in a multi speaker setup.
For a 5.1 system, the reference level for each speaker is 78 dBC, so that they sum up to 85 dBC.

@dasdoing Thank you for the link to the ITU document!

I realized that Audacity has a loudness normalization effect.
Pink noise at -14 LUFS has an RMS of -13.5 dB, so not far off. (RMS 0 dBFS for FS sine.)

Here is a table of loudness targets of streaming providers:
https://www.masteringthemix.com/blo...dio-for-soundcloud-itunes-spotify-and-youtube
Some go up to -9 LUFS or even higher.

I have no streaming subscription.
But I wonder if there is a way to measure LUFS of a soundtrack played via streaming.
As I understand it, it is not possible to download a soundtrack from a streaming service (if so, then only encrypted).
So, it would not be possible to analyze the soundtrack with some software.
But maybe there is a way?

 

[dasdoing]

I have no streaming subscription.
But I wonder if there is a way to measure LUFS of a soundtrack played via streaming.
As I understand it, it is not possible to download a soundtrack from a streaming service (if so, then only encrypted).
So, it would not be possible to analyze the soundtrack with some software.
But maybe there is a way?

no clue how to do it in Windows, but in Linux I simply can conect the streaming output to the LUFS meter input.
you could also record the stream in a DAW, but that would be a crime hehe

 

[Aerith Gainsborough]

so that they sum up to 85 dBC.

I realized that Audacity has a loudness normalization effect.
Pink noise at -14 LUFS has an RMS of -13.5 dB, so not far off. (RMS 0 dBFS for FS sine.)

Umm... guys?
Small problem here: I generated a file with -13.5dB RMS in Audacity (thanks for the tip!) and measured it's output.
At the maximum volume of my system I only get to ... erm... 66dB (C). (Digital volume is at ~ -12dB FS, so not much headroom there)

Are you sure that I'd need to calibrate that one to 85dB? Music would be insanely loud, basically +19 dB to the highest value I'd ever use.

Or did I misunderstand something here?

 

[dasdoing]

Umm... guys?Small problem here: I generated a file with -13.5dB RMS in Audacity (thanks for the tip!) and measured it's output.At the maximum volume of my system I only get to ... erm... 66dB (C). (Digital volume is at ~ -12dB FS, so not much headroom there)Are you sure that I'd need to calibrate that one to 85dB? Music would be insanely loud, basically +19 dB to the highest value I'd ever use.Or did I misunderstand something here?View attachment 101326

damn, I never meant to create a new standard for calibration lol....I just mentioned what I have done.

I just have opened a pink noise in Audacity and used the above mentioned loudness normalization effect to bring it to -14LUFS. this file played back realy produces -14 LUFS.

now, if you calibrate to this, any music that meassure integrated LUFS above that -14 level is obviously too loud. if one calibrates to that level it is a must to have -14 LUFS sources.
I just meassured Mendes/Cabello - Señorita MP3 and it has -8.1 integrated LUFS...so on Spotify it would be attenuated by 5.9dB!!!....and that is roughly what I do when playing back that MP3 (roughly beacuse I monitor "momentary loudness")

 

[dasdoing]

I do so because I use SPotify all the time.
If one manly uses CDs as sources, normalizaing everything to -14LUFS is probably overkill......-14 was chosen because it is possible to get every existing genre to the same loudness (*). if you take only pop music and/or electronic music you could get away with a much higher value.

(*) if you try to bring a dynamic track to those -8.1 LUFS of Señorita it will clip

If I hade a library and only listened to that I would probably scan the whole thing and normalize everything to the highest possible integrated LUFS level > and then calibrate to this level

 

[playmusic]

Small problem here: I generated a file with -13.5dB RMS in Audacity (thanks for the tip!) and measured it's output.
At the maximum volume of my system I only get to ... erm... 66dB (C). (Digital volume is at ~ -12dB FS, so not much headroom there)

When you generate pink noise in Audacity for -14 LUFS you can do as @dasdoing described or you can create it directly with an RMS of -16.5 dB.
The reason I wrote -13.5 dB was that it was for the AES standard which sets RMS of a FS sine to 0 dBFS. But Audacity sets RMS of a FS square to 0 dBFS. So, there is a difference of 3 dB. Something to keep in mind when using different audio programs.

So, the -14 LUFS would sound even 3 dB lower on your system.

How did you determine that the test signal has 66 dB SPL (C) on your system?
Did you measure it with a physical sound meter in your room? Or did you calculate it?

And why did you set digital volume to -12 dBFS?
Is it because of intersample clipping?

Unless you have a good reason to do otherwise, you might keep digital volumes in the chain at 0 dB.

If you have checked that your full chain after the audio source is really 24 bit then digital volume control should have no effect.
(For clarity: I think that 16 bit audio source is fully sufficient. There are no gains with 24 bit for consumers. For recording/mixing, 24 bit has a reason of course.)
But for simplicity, I keep my digital volumes at 0 dB.

Anyway, I listen with loudspeakers a lot lower than 85 dB SPL usually.