? Audyssey DEQ RLO isn't for accomodating different speakers. It's for accomodating different production or output styles (different levels of average loudness). If you're playing mostly hip hop you'll use a different RLO than if you mainly listen to chamber music.
-
Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Should you use Fletcher-Munson loudness compensation?
- Thread starter dshreter
- Start date
? Audyssey DEQ RLO isn't for accomodating different speakers. It's for accomodating different production or output styles (different levels of average loudness). If you're playing mostly hip hop you'll use a different RLO than if you mainly listen to chamber music.
Yep, I've heard something along that line myself, IMHO the practice is BS and at the minimum should be switchable.
It noticeable upsets that F / R balance, is not part of the F-M research, and I know of no one else in the industry doing something similar with multich loudness compensation.
For sure, whats the big deal. One of the coolest features of the $20 Editor app was being able to switch off that midrange dip.
Another Audyssey idea that was simply a preference thing that should have been users choice from day 1.
- Joined
- May 22, 2020
- Messages
- 617
- Likes
- 288
Divinity runs on an ipad?
I wrote two scripts (works on any browser, no installation needed) covering both papers. The first one generates Equal Loudness Contours from 40 to 90 in steps of 5 (customizable in any text editor) according to ISO226 2003 standard.
The second script applies Steven's proposed revisions in the second paper (to my best understanding of it) and generates directly the differences from reference 40 phons. Similar target curves can also be generated from the ISO226 curves by ¦A¦/¦B¦ trace arithmetic in REW (align all curves SPL to 0dB at 1000Hz, 0 octaves to get comparable division results).
You can drag and drop both zip contents directly onto REW.
The 85dB Steven curve fits perfectly as a target curve to my speakers steady state response measured at 85dB. Standard ISO curves are a bit too steep.
Last edited:
The FM curves are interesting in that they just show how futile the idea of sound fidelity is. FM compensation sounds good, that's why the Sonos systems sound so nice I think. But look at the range of tonal compensation! At 70db, it's something like a 20db compensation from whatever our reference was.
So not only are our recordings created with who knows what microphones, with their arbitrary spatial and tonal response, but they are mixed at some unknown level, potentially different at levels at different times, each having a huge effect on the tonal balance.
Old receivers had a 'loudness' button but in 2024 I'd love to see a DAC that has two knobs, a FM compensation knob and a volume knob. Throw in a stereo width knob while we're at it.
Of course we'll never see that, DAC manufacturers are too busy chasing that next dB of snr rather than anything that makes music sound good.
So not only are our recordings created with who knows what microphones, with their arbitrary spatial and tonal response, but they are mixed at some unknown level, potentially different at levels at different times, each having a huge effect on the tonal balance.
Old receivers had a 'loudness' button but in 2024 I'd love to see a DAC that has two knobs, a FM compensation knob and a volume knob. Throw in a stereo width knob while we're at it.
Of course we'll never see that, DAC manufacturers are too busy chasing that next dB of snr rather than anything that makes music sound good.
ZolaIII
Master Contributor
- Joined
- Jul 28, 2019
- Messages
- 5,190
- Likes
- 3,047
Well thers software which can do it (ISO 226) without knobs of course and you can do normalisation on material side hopefully EBU R128 and for free if you are not very picky and with PEQ and convolution (EQ-APO). Or you can do it with self filters pretty good. Now days we are starting to have dongle alike pocket DAC's with full PEQ support and more than fine and audible transparent performance. Future more you have simple apps that does it all like Wavelet (Android only and paid for ISO 226) and Web based (browser) one's for producing EQ's like Auto-EQ. So it's more reachable than ever but people's habits didn't exactly change nor the most of them goes that mile even if it doesn't costs them anything. Just to mention also wasted huge processing power (like server machine a decade ago) on the palm of your hand. Just don't expect from industry to resolve it, they live from telling lies.
Absolutely - which is why a subjective, manual setting of the reference level is required....
It definitely shouldn't be just an on/off.
Thank you for this. I have two rookie questions. Would it be difficult to update it to the ISO226 2023 standard? And even if yet, would you still recommend the Steven curve?
Also is there an obvious way to load these into the Wavelet android app? I have tried to convert it, am pretty sure I butchered it.
The 2023 ISO paper is not available publicly but if anyone can share it here, I can happily implement. SInce I haven't seen it, I don't know if Steven's corrections are still better or not. I am on IOS and PC WIndows and haven't heard about the Wavelet app
AncientMariner
Member
- Joined
- Mar 13, 2024
- Messages
- 74
- Likes
- 55
As a big fan of loudness, my take on this subject:
It would be very wrong to tune speakers in any way to the shape of these curves. These curves represents how we listen to the sound, not how it's played back. If a flat microphone recorded the sound and it's played back on a flat speaker, these curves are baked into the recording, with the only variable being the recording/playback level.
In my opinion, the only meaningful info that can be detracted from the curves is, what everyone knows intuitively already: That we have a harder time hearing bass and treble at lower playback volume.
With the variability in HTRF etc., it would probably even be pseudo-science to even try to extract any loudness targets from the curves as well. I would not even recommend normalized curves as this is not how your brain expects to hear music a lower levels. Most of the time if you want to turn down the volume, it's because you want to concentrate on something else (such as writing a post on ASR). Your brain expects a reduction in certain frequencies.
The purpose of loudness is simply to create a pleasant listening experience at lower volumes, while still giving your brain the appropriate signals that the volume was indeed reduced. If you want do concentrate on the music, the simple answer is turning the volume up!
Thus the right way about it is to have a speaker that plays flat at louder volume levels, and for those with DSP available apply a loudness filter at lower volumes. For example CamillaDSP has great and configurable volume dependant loudness filter available.
The way I do it is to find the loudest level I would want to listen to the speaker and using gain to lock that as max volume in CamillaDSP. Then I turn the volume down to find the lowest level where I'm still satisfied with the bass and treble. Below that level I let CamillaDSP apply a loudness filter that gradually boosts the bass and treble over the subsequent 20dB reduction in volume (a 6dB boost works well for me on two different sets of speakers). Of course, I can't prove this is scientifically correct, but it makes the music very enjoyable to listen to at lower volume where it would otherwise sound dull.
It would be very wrong to tune speakers in any way to the shape of these curves. These curves represents how we listen to the sound, not how it's played back. If a flat microphone recorded the sound and it's played back on a flat speaker, these curves are baked into the recording, with the only variable being the recording/playback level.
In my opinion, the only meaningful info that can be detracted from the curves is, what everyone knows intuitively already: That we have a harder time hearing bass and treble at lower playback volume.
With the variability in HTRF etc., it would probably even be pseudo-science to even try to extract any loudness targets from the curves as well. I would not even recommend normalized curves as this is not how your brain expects to hear music a lower levels. Most of the time if you want to turn down the volume, it's because you want to concentrate on something else (such as writing a post on ASR). Your brain expects a reduction in certain frequencies.
The purpose of loudness is simply to create a pleasant listening experience at lower volumes, while still giving your brain the appropriate signals that the volume was indeed reduced. If you want do concentrate on the music, the simple answer is turning the volume up!
Thus the right way about it is to have a speaker that plays flat at louder volume levels, and for those with DSP available apply a loudness filter at lower volumes. For example CamillaDSP has great and configurable volume dependant loudness filter available.
The way I do it is to find the loudest level I would want to listen to the speaker and using gain to lock that as max volume in CamillaDSP. Then I turn the volume down to find the lowest level where I'm still satisfied with the bass and treble. Below that level I let CamillaDSP apply a loudness filter that gradually boosts the bass and treble over the subsequent 20dB reduction in volume (a 6dB boost works well for me on two different sets of speakers). Of course, I can't prove this is scientifically correct, but it makes the music very enjoyable to listen to at lower volume where it would otherwise sound dull.
AncientMariner
Member
- Joined
- Mar 13, 2024
- Messages
- 74
- Likes
- 55
Perhaps I didn't expres myself clearly. No, loudness compensation is great, and as I started saying I'm a big fan.
What I adviced against is using equal loudness contours (Fletcher-Munson curves) as basis for loudness compensation.
- Joined
- Jun 24, 2024
- Messages
- 1,147
- Likes
- 1,607
I see. So if I'm reading it right now, you're saying you apply a gradually increasing boost to the bass and treble that maxes out at 6dB once you hit a certain volume level which you decide upon by ear?
AncientMariner
Member
- Joined
- Mar 13, 2024
- Messages
- 74
- Likes
- 55
That's what works for me, yes. But it's subjective. What I would advice is to simply adjust it to your liking. If you're "critically listening" (not a fan of that term), i.e. concentrating on the music and want an accurate representation, you're likely to listen at a louder level anyway.
You can try to make a filter like described by solderdude on page 2:
Should you use Fletcher-Munson loudness compensation?
I know many are familiar with the Fletcher-Munson or equal loudness curves. In a nutshell: “The Fletcher Munson Curve contains a set of graphs that show that when you listen to music at a lower volume, mid-range frequencies will be sound more prominent, whereas high and low frequencies will be...
audiosciencereview.com
It sounds really really bad. It's muddy, dark, and lacks treble. There is no clarity. Vocals are mumbled.
Compare that with the filter in the second screenshot. Sound is clear and airy. Vocals are crystal clear as they would be if listening a much louder level.
My own personal conclusion is that equal loudness contours only tell how a person perceives the level of one frequency vs another in isolation (which is how the tests were conducted). It's not really valid to how we perceive complex sounds such as music. Our brain is a powerful EQ on its own and has expectations how combinations of sounds should sound like.
Edited: As a sentence accidentally slipped out.
Attachments
Last edited:
I think it would make some sense that a more generic, gentle, low Q filter would be less invasive than following the loudness curves exactly. It sort of makes sense not to try to nail the wiggle in the treble region, for example.
AncientMariner
Member
- Joined
- Mar 13, 2024
- Messages
- 74
- Likes
- 55
I thought of an example to illustrate that I don't mean that in a mumbo jumbo way:
Imagine you listen to a girl singing who is standing in front of you. You now move 20 meters away from her, and the volume of her voice is reduced. According to the research done by Fletcher-Munson, the lower frequencies of her voice will have dropped off by some dB to your ears.
But your brain will compensate and her voice will sound the same to you just at a lower level. You're not going to think "wow, her voice sounds really thin now". You brain EQ'ed the missing dB.
Now you listen instead to a girl singing through your speakers. You listen at 80dB and it's like she is standing in front of you. Now you reduced the volume by 20dB. If you use Fletcher-Munson curves to compensate for the lower volume, you boost the lower frequencies of her voice, and all of a sudden her voice sounds muddy and dark. Because you brain will not accept this is the tonality of her voice at this volume level/virtual distance.
This is why I don't think it's correct to compensate using Fletcher-Munson curves.
We can, however, boost the overtones/treble to add clarity, and the sound below 200Hz to add some of the tactility that was lost due to the lower volume because it won't impact the tonality of vocals and instruments that much.
I think the objective with loudness should be to obtain a pleasing sound, which may include tricking our brain. The accurate representation is to play back the music at the same level it was recorded.
Similar threads
- Locked
