Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: ISO 226:2003 based loudness correction (Read 20384 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

ISO 226:2003 based loudness correction

Reply #25
What's your reason for using Ozone instead built-in parametric equalizer, maybe lower number of bands provided in CEP/Au, or something else?


Foobar's EQ ends at 55 kHz at the bottom. Also missing exact meters and fixed spacing of bands didn't allow me to replicate the curves as exactly as I wanted. I am also no big fan of graphic equalizers in general. They always either don't fit the problem sufficiently or implement basically very simple things very complicated.

ISO 226:2003 based loudness correction

Reply #26
Nice work. I went against your advice, and did some quick listening with MP3s of Portishead; Mysterons and Sour Times, specifically.

I preferred the -20dB version for very low level playback, and didn't find much use for the subtly different -10dB, as I generally don't notice any disturbing low/high attenuation when playing at the levels it was apparently meant for.

The loudness button on my amp seems smoother than the convolver wavs, provides a greater low boost across a greater range, and also amplifies the high ranges significantly more.

ISO 226:2003 based loudness correction

Reply #27
Foobar's EQ...

I was talking about Audition built-in parametric eq
Isn't -20 dB corrected too attenuated compared to same alternative (by looking in CEP frequency spectrum)?

 

ISO 226:2003 based loudness correction

Reply #28
Oh, I guess that would also have been fine. I'm more familiar with Ozone's UI and didn't have to research first if it's any good.

ISO 226:2003 based loudness correction

Reply #29
For an in-depth discussion of loudness compensation, including suggested circuits, see this thread over at diyaudio:
http://www.diyaudio.com/forums/solid-state...ss-control.html

The practical side of that thread is aimed at processing in the analogue domain. I hadn't considered "doing it digital" as this thread has suggested.

By the way, the Matlab function is easily converted to other languages - I ported it to REXX without any problems. It should be just as easy to port it to VB, C++, Perl etc.
Regards,
   Don Hills
"People hear what they see." - Doris Day

ISO 226:2003 based loudness correction

Reply #30
Isn't -20 dB corrected too attenuated compared to same alternative (by looking in CEP frequency spectrum)?


Which alternative? The -20 dB curve matches the iso226(83) - iso226(20) function exactly and was made by hand without any personal tweaking. -10 dB has been generated from that with Ozone's "amount" function, i.e. I left all settings untouched and reduced the EQ's "amount" value until the curve matched the new iso function within +/- 0.2 dB again. It worked without further adjustments. Condition is that Ozone is really able to output what it displays, but as yet I really haven't had a reason that call that into question.

ISO 226:2003 based loudness correction

Reply #31
"-20 dB Loudness Corrected.wav" and "-20 dB Loudness Corrected - Alternative.wav" when looked in frequency spectrum: http://img215.imageshack.us/img215/2696/clips.png

I was just asking as I don't quite understand how this works, because when impulse is applied to signal they are both correcting as it should and "auto level" is disabled in convolver

ISO 226:2003 based loudness correction

Reply #32
Please double check. The plots you have uploaded do indeed not look as they should. But I just did the same analysis in Audition and it looks fine.

ISO 226:2003 based loudness correction

Reply #33
Yeah, it's mistake at my end: the second screenshot is with response from cursor position and not entire impulse (I didn't pressed "scan"). Because impulse is at the beginning of waveform the frequency response looks similar in shape as for the entire impulse, and that's why I haven't noticed that whole impulse is not selected. Apologies for inconvenience

ISO 226:2003 based loudness correction

Reply #34
I really don't see your problem.

It seems to me I haven’t any.
Maybe it is just a different use of words...

Yes, very different.
I call "reference" what you seem to call "natural".

From your reasoning, I see absolutely not. Let’s see:
What do you guys think would be the best reference curve? What's a typical SPL of a rather loud but still comfortable listening session?

The crucial difference in the use of words in our specific case, is that the only reference SPL for equal loudness compensation applications in all my reasonings is the SPL natural for the natural sound source, listened to electro-acoustically at any moment. And not a “rather loud”, or for “comfortable listening session” etc.

What can be a “typical SPL” common for a flute, with its typical SPL of about 60…65 dB, and for a forte-fortissimo of a big symphony orchestra with a chorus, with its typical peak SPL of about 105…110 dB?

I also do not agree that the only significant range is 20...700 Hz. For example, why omit a over 2 dB difference at 12.5 kHz for an 18 dB attenuation?

I don’t know what you mean by 18 dB attenuation, but analyzing the equal loudness curves version ISO 226: 2003
http://www.lindos.co.uk/cgi-bin/FlexiData....=full&id=17
I don’t see any difference in the ordinates values of the curves in the 12 kHz zone within a reasonable hearing range of 40 to 100 phon.
Why not just model perfectly? A four band parametric EQ is sufficient to get an accuracy of +/- .5 dB with respect to the mathematically derived curve.

In my opinion, by the simple reason that there cannot be mathematically derived curves for equal loudness compensation, because nobody can know the “mathematically” accurate value of the SPL created by the natural sound source at the sound picking up place, and therefore the “mathematically” accurate difference of this value and the value of the SPL created by this natural sound source listened to electro-acoustically.

Hence it follows the point that I’ve mentioned in my previous post:
The only criterion applicable is the listener’s preference.

I'm wondering if Fedot means that the pre- and post-ReplayGain values for each track (or at least the difference between them) would be required to calculate the natural and target loudness levels respectively?

No, I didn’t mean this, because to me, no calculation for these purposes is possible.
After all, a track that's intended to be quieter originally should still sound quieter in terms of tonal quality after correction if the relative apparent loudness level of the track is to be representative of reality.

I haven’t understood what you meant by this.

So I can only say that a sound source listened to electro-acoustically may only be representative of reality in its general quality, if it sounds at its natural SPL. Any sound source listened to “louder” or “quieter” than natural inevitably sounds unnaturally, “artificially” by definition.

And equal loudness compensation is inevitably an artificial tool intended for to artificially communicate to a sound source sounding unnaturally when listened to “louder” or “quieter” than natural, a tonal balance the listener is accustomed to at its “natural” SPL.

Under the precondition that the electro-acoustical system is able to reproduce linearly the whole frequency range necessary for the given sound source.
From the experiments I carried out when designing an analogue loudness control for a home-made preamp many years ago, I'd agree with you that a degree of correction is also desirable at the higher end of the audio spectrum.

In principle, as the “equal-loudness contours” cited above show, high frequencies don’t need a tangible compensation.

But the equal loudness compensation problem is much more complicated than it seems to be.

First, by the fact that the high frequencies having normally much lower levels than others, disappear quickly “sinking” in the ambient noise (masked by ambient noise) with decrease of general listening SPL level. That makes the listener boost the HF for to make them audible.

Furthermore, many natural sound sources having a very large natural dynamic range (up to 80 dB in extreme cases of forte-fortissimo of a big symphony orchestra with a choir and soloists) can’t be listened to anywhere except big theatre or concert halls, a fortiori, electro-acoustically attenuated, where most “quiet” passages will simply be lost!

The reason, by the way, to always compress the dynamic range of such natural sound sources, and very considerably, from 60…80 dB down to 40…50 dB on all analogue recording mediums (tape, records, radio), due to their limited S/N ratio.

Such a compression changing very radically the “natural” proportions between the quietest and the loudest fragments, and therefore, making meaningless any “mathematical calculations” of the spectral balance of programs listened to electro-acoustically. Thus making me simply return to my initial opinion that the only criterion for equal loudness compensation is the listener’s individual experience and his individual preferences. And undoubtedly, the ambient conditions of listening.

ISO 226:2003 based loudness correction

Reply #35
What can be a “typical SPL” common for a flute, with its typical SPL of about 60…65 dB, and for a forte-fortissimo of a big symphony orchestra with a chorus, with its typical peak SPL of about 105…110 dB?


I think I have finally understood your criticism, but I must still reject it. Of course a flute has a different SPL than the fortissimo of a whole orchestra at any given (linear) volume control setting. But that's not relevant!

At the time of mastering (or even already when recording) all these instruments and their respective volume levels are put into a common context. This context is created by the recording and/or mastering engineer at about 83-85 dB SPL in front of a monitoring setup with flat frequency response. This is exactly the same context you want at playback time! And it will already have undergone the first step of human ear loudness contour correction by the person having done the mix.

And your pair of speakers at home will have exactly one volume setting coming closest to that monitoring situation (not necessarily at the exact same SPL setting) with respect to perceived frequency response. The exact flute SPL within the reproduced context, in comparison to a tenor, is irrelevant. You want to preserve their relative loudness to each other as it has been defined at 83-85 dB SPL, flat FR, by the engineer.

But as soon as your overall playback volume differs too much from the "reference" level, where your speakers produce the closest approximation to the mixing setup, relative loudnesses, as we perceive them, start to get altered. This alteration is highly correlated to the differences of loudness contours as defined in ISO 226:2003. To restore a perceived loudness contour at a level n dB below the level, where your speakers produce the closest approximation to a flat FR, 83-85 dB studio setup, the loudness contour resulting from iso226(83) - iso226(83 - n) will result in a very accurate correction curve able to restore the intended context. As I have also already written, if you do the math, the choice of a specific reference level is not that significant iso226(60) - iso226(60 - n) delivers almost exactly the same curve. n, what I have also called "attenuation" is the only significant variable.


I also do not agree that the only significant range is 20...700 Hz. For example, why omit a over 2 dB difference at 12.5 kHz for an 18 dB attenuation?

I don’t know what you mean by 18 dB attenuation, but analyzing the equal loudness curves version ISO 226: 2003
http://www.lindos.co.uk/cgi-bin/FlexiData....=full&id=17
I don’t see any difference in the ordinates values of the curves in the 12 kHz zone within a reasonable hearing range of 40 to 100 phon.


I also don't see it by just looking at the curves. Get the raw data, do the math and you will get exactly the results at 12.5 kHz, that I have posted. "18 dB attenuation" is n from above, the difference in dB at 1 kHz between two ISO 226 curves.

ISO 226:2003 based loudness correction

Reply #36
It's a nice effect googlebot - certainly nicer than the one I had on a very old stereo.

The proof of the pudding would be having it tied to a fully variable volume control, and seeing how natural (or not) it sounded as that was changed.

I don't know if, to my ears, even a "perfect" implementation would sound "natural". In real life if someone plays quieter, or if someone is playing further away, the equal loudness curves of my ears aren't compensated for - I hear them in full effect. Removing them isn't natural. But turning a symphony orchestra down to 60dB isn't natural either!

If you're going to compensate for lower listening levels, IMO you also need some DRC to bring the previously quiet and now completely inaudible sounds back into the audible range. Dolby Volume does this quite unobtrusively - not sure if there's any EQ in there though.

Cheers,
David.

ISO 226:2003 based loudness correction

Reply #37
Yes, at extreme settings the "naturalness" is indeed debatable. Quiet things have a "signature" of being quiet and applying a large amount of frequency domain correction delivers an odd mix, at least one that one isn't accustomed to. I underwent three stages with my first -24 dB trial: 1. Eww, what's that? 2. Thorough comparison uncorrected vs. corrected vs. loud. Result: corrected, even though sounding unfamiliar, had a better representation of relative instrument volumes compared to uncorrected. So I made the -20 dB patch to reduce the effect a little and settled for that. 3. Couldn't get over the impression of boomy-ness for some tracks. With over 10 dB boost in the 20 Hz region I guess my speakers were also starting to contribute some distortion at the lower end. In response to that I settled for the -10 dB patch and as yet I'm really satisfied with it. I think about -12 dB could be about my personal sweet spot. I'll make that when I find the time. A variable control would surely be the best solution.

I really started this project to get around DRC. Finding a one size fits all parameter-set for attack and release times is really hard. I can live with quiet parts being to quiet if I turn down the volume control. Really dynamic music just isn't suited for very quiet listening sessions in my opinion and I don't know if I need to fix that.

When I get the chance, I will have a look at Dolby Volume. Didn't even know that it existed. It should be somewhat limited by its dynamic realtime nature, though.