The loudness war....

[mpmarino]

Yes, join the club! What we are discovering in our old age is that the originators and first providers CARE about quality: those who come later care about PROFIT.

But be reasonable! SPECIFY what you want (and extimate the market for that) before complaining.
Perfection is not a spec. Neither is "like the good old days".
A decent spec includes HOW it will be measured instrumentally
John

I'm not that old - the future looks grim

Your point is well taken. I'm sure they used 'some' reference back then. I think that using the 'reference' used back then combined with the s/n of modern recording technology we would have some pretty listenable material. Your right tho, you probably can't properly moan without doing the numbers.

 

[AndrewT]

.........The 120 db is a statement about the ratio of the loudest sound to softest sound reproducible.
But the sound AVAILABLE to record is:
Loudest Full orchestra 100 db
Softest 35 db of quiet country living room.
Ratio 65 db

I don't think that is right.
the peak levels in live music are higher than 100db. I think they can be much higher.
I was at a school African drumming band exhibition a few weeks back, just 19 drums in the ensemble. It would be my dream to be able to reproduce that in my listening room, but only a dream.
I doubt my best sytem could reach those peak levels and if they could get near the distortion on the heard signal would make it obvious that it was not the real thing. The crack, off some of the snares, was alarming.

However, even if we could reproduce at the highest levels with all it's attendant problems, I think the more difficult part is at the quiet end.

If the average level we choose to listen at is 80db leaving room for +20db to +30db peaks then at the quiest end when some of the smaller signals are at a level of say 40db to 50db below the average, we are asking the CD to send a signal that is now about (30+50db) 80db below 0dbFS. In this signal are the harmonics of the instrument. We all know it is the proportions of the harmonics that differentiate between brass and woodwind and steel guitar and gut and .. and .. etc.
What levels are these harmonics below the main quiet signal?
What if the ratio between the harmonics is not preserved?

What is the signal like that comes off a CD when operating at the lowest levels?
If all the bits say off then we have nothing, well almost nothing, the noise if not dithered is at -96.3db if the DAC is perfect and mute has not operated. If the LSB turns on we have a voltage level of -90.3db relative to 0dbFS. if the second bit turns on we have -84.3db (did i get that bit right?). These three lowest levels are supposed to accurately reflect the ratio between nuances of different instruments and our ears/brain try to interpret the reproduced signal to recognise what we are supposed to be hearing.
Basically "music" at the -80 to -96db level is unrecognisable, in a 16bit recording system. It is completely incapable of maintaining the ratio of harmonics of quiet music signals.
Now apply some compression and let the quietest harmonic signal move up to -70db to -80db, then there is some hope that what comes out of the CD player will be just about recognisable due the discrete steps between each of the bits now being fine enough to preserve some of that important harmonic structure.
Forget about 96.3db SNR it does not apply to music levels. Accept that the range of music off a 16bit system is only about 70db to 75db. Not nearly good enough.

Now if we could get 20bit, that would be a much better target for achieving good music reproduction for a decent dynamic range. 22bit is obviously even better. We have not reached 24bit yet. I wonder how long till the Technology catches up?

 

[Empee]

MrWhy,

we're not complaining of the perfomance of the microphones on the air being non-lineair....


we're complaining of the (over-)compression used on the complete work, to better suit bad repreduction devices used by 'the masses'

 

[mrwhy]

Microphones are tremendously improved since those days.
What has gone downhill the music that purchasers want to PAY FOR.

 

[mrwhy]

What is the signal like that comes off a CD when operating at the lowest levels?

This hits the nail on the head.
The answer is VERY Bity
And SUCHA pity!

The smallest digital number is zero and one bit above that is an infinite number of db more! (actually it is "two-complement of summat")

The harmonics of many instruments (e.g guitar) are stronger than the fundamental (our ears put in in anyway - even when it is completely absent).
But lets say we are listening to middle C and want to hear all its harmonics up to 20 Khz.
This is impossible!
Why?
Well how loud shall we guess the 79th harmonic is likely to be (or the 80th if you think that stronger)
If you wanna guess - be my guest.
My guess is average about 6 db per harmonic over so many harmonics. Too many db for a cd.

So look at it from the other end.
At what loudness level - expressed as a digital number - is a change of one bit in the units column 1%?
Would be when the digital number changes from 200 to 201 (power going as the square)
In other words of the 120db claimed only 1/200th is available to give us sound where we get the harmonics without distortion if they are louder than 1% of the fundamental
So 120 db minus 23 db (1/200) is 97 db.

But what worries me is with the drums in one channel at threshold of pain you MIGHT hear (or dream of hearing) their hundredth harmonic in THAT channel if itwere not compressed or peak limited. But what about the picolo in the other chanel. We may HATE compressors but it stops the drum cross-modulating the picolo (or even pulse-modulating it off/on)
And remember, in all this, that the timbre of sound only sounds right when heard it at the normal level - NOT at 120 db.
So if we DO have an 80 db sound its harmonics will distorted more than 1% at their level of 80-23 = 57 db.

So though ONE link in the reproduction chain (the cd) iis claiming 120 db, there is at least one link (the only way of using it because it is digital!) that reduces the 120 db to 57 db!
John

 

[mrwhy]

I'm not that old - the future looks grim

Your point is well taken. I'm sure they used 'some' reference back then. I think that using the 'reference' used back then combined with the s/n of modern recording technology we would have some pretty listenable material. Your right tho, you probably can't properly moan without doing the numbers.

So what one of you fine young people should do is
GET THAT SPEC
Set up a company to market cds that meet it!
(Like Deuche Gramophon did years back)

You would need permission to ave your own recording equipment present - but I am sure they would not miss the tiny market you would thus rob them of.
John

 

[mrwhy]

Digital techniques inherently distort!
Put me right about this, guys:
If you look at the actual voltage waveforms during a live performance, you see positive and negative excursions about zero.
They are NOT equal. On some microphones the positive transient peaks are greatest – on others the negative. Or maybe it is where the performer stands.

So if we are going to faithfully record this waveform we have to decide on a number to represent undisturbed air pressure. So with 16 bit recording that would be the number 32,000 (in binary). So our peak-to-peak range is 8 bits positive to 8 bits negative – total 16 bits.
We do not know if the positive peaks will exceed the negative ones, or vice versa, so we must allow say 6 db to allow for this ignorance.
So our signal excursion will be between the digital numbers 16,000/4 and minus 16,000/4.
In this range we represent a sinewave by a staircase of harmonics!
(Have to now continue in the next post!)

 

[mrwhy]

(continued)
The step size, at the max volume we dare record is one bit.
So if the volts is now represented by the number 8000 (peak volume!) what is the % change that is the SMALLEST the digital numbers are capable of expressing?
The analogue waveform is of course SMOOTHEST at its sinewave peak. But digital coding represents this as a change from 4000 to 3999.
That is a jump of 1 in 4000 or 1/40 of 1% error.
So for the loudest sinewaves the digitally-generated harmonics may not be objectionable.
But if we want (dream of) a dynamic range of 80 db we must think how the digital process messes up a signal ten power 4 times weaker in volts.
The digital distortion is now 250%. WOW!
The digital process is chopping off just the peaks of the waves and replacing the rest with silence. The signal crossover (for weak signals) is severe crossover distortion jumps!
Have I got this even roughly right?
I am sure you will tell me!
John

 

[AndrewT]

Hi,
I am not competent to describe the digital process for you, but I can confirm that the encoding system is clever enough to do 16bit accuracy for level AND also to define whether each of these levels are above or below the zero voltage level without losing any bit accuracy.

 

[jcx]

not quite right guys, try a little reading:

http://www.meridian-audio.com/w_paper/Coding2.PDF

for best results skip to end to look over graphs/illustrations, jump forward forward for text/discussion - repeat cycle

advanced noise shaped dither actually delivers better than 96 dB perceptually weighted S/N and "perfect" linearity with RedBook CD

http://www.digido.com/bob-katz/dither.html

http://www.users.qwest.net/~volt42/cadenzarecording/DitherExplained.pdf

the last is a little simplistic

lots more dither, noise shaping info on the web

 

[mpmarino]

Modern studio recorders use multi-band compressors that work on individual frequency bands, without knowing the settings, it would be just about impossible to undo.

Al,

Are you saying that this mess is created at the time of recording?!

I had always imagined that there was an uncompressed master sitting in an archive somewhere. If your above statement is as I understand it, there is little hope of recovering decades of good work. I know there would be few buyers, I'm speaking hypothetically....

 

[AndrewT]

advanced noise shaped dither actually delivers better than 96 dB perceptually weighted S/N and "perfect" linearity with RedBook CD

I don't think that noise is the problem nor is it lack of linearity. It's the poor attempt of the analogue to digital (ADC) sampling of low level SIGNALS that seems to be at the root of one of the problems namely reproducing realistic music when the levels go down a bit. The other problem is over recording that seems to have become much more prevalent in the last decade or so. The pair of problems taken together show that 16bit is incapable of accurately reproducing the dynamic range in live music.
It does not matter whether our systems can or cannot reproduce at 100db or 110db or even 120db. The recording system has to be compromised (compressed) somewhere to fit natural dynamic range into 16bit.
I suppose that's why they invented HDCD coding to tryand get some extra low level information into the ADC and DAC chain. Is it capable of 20bit?

They do something similar with FM radio in the UK.
They distribute 14bit audio on a 10bit data stream, but use coding to move the data stream up and down the required level band and decode it back to full 14bit accuracy at the transmitter. When done properly, with minimal EQ at the transmitter, it sounds wonderful.

 

[jcx]

my you're a quick reader

really read the Stuart paper - no one today is using 16bit master recording, cheap monlithic 24/96K ADCs are near 120 dB s/n and have differntial linearity below that, I expect the very best multi-chip/discrete designs do even better

the question of why "loudness war" levels of compression on CD recordings has nothing to do with high bit resolution, high sample rate mastered material properly resampled and dithered to best use RedBook CD resolution

 

[AndrewT]

But we still have the 16bit data on CD.
and MP3 is an encoded version of this same 16bit data.
we can never overcome the limitations of 16bit as long as were are forced to use CD or the existing MP3.

Now if we were to use a new version of MPx that was an encoded version of 192ks/S and 24bit and squeeze it into the same data bandwidth as 16bit CD, then we will have achieved something worth reproducing.

DVD video is doing something that moves us part way along that route.
HDCD also attempted to move us along a similar improvement route.

It matters not that they record at the technological achievable limits if we are strangled by the 16bit mediums we are currently stuck with.
This will always show up in any loudness test whether our average level is 70db or 80db or 90db. Only if we turn down the system to significantly less than 70db of average level does the inaccuracy of 16bit coding get buried in the ambient noise, or listen in a noisy environment. Well, listen at comfortable levels in a moving automobile and see how much the high ambient noise levels hide the inadequacies of 16bit. I have heard that some people even accept MP3 in the car

Has anyone got comparative data on the differing audio standards between DVD video and HD DVD video either in two channel mode or multi-channel mode?

 

[pinkmouse]

Al,

Are you saying that this mess is created at the time of recording?!

I had always imagined that there was an uncompressed master sitting in an archive somewhere. If your above statement is as I understand it, there is little hope of recovering decades of good work. I know there would be few buyers, I'm speaking hypothetically....

Yes and no.

Multi-band compression is most frequently used in the mastering process, but is also used in the recording process for high dynamic range signals like voice or drums.

However, what is worrying, is that less and less master "tapes" are now kept. With the advance of digital and hard drive recording, there are no masters. Quite often with smaller bands or newer artists, there is no real permanent archive. The days of wandering back from the studio with a reel of 24 track are over, now they walk away with a premixed CD, or the final studio mix gets fired via the internet directly over to the mastering/pressing plant...

 

[phase_accurate]

So with 16 bit recording that would be the number 32,000 (in binary). So our peak-to-peak range is 8 bits positive to 8 bits negative – total 16 bits.

15 bits to the positive and 15 bits to the negative would be a better way to look at it.

While it is agreed that the lowest part of the dynamic range of a CD isn't able to represent any signal accurately it does still not make sense to compress CDs such that only the uppermost 10 dBs of the system dynamic range is used.

Regards

Charles

 

[jcx]

...it is agreed that the lowest part of the dynamic range of a CD isn't able to represent any signal accurately...

Regards

Charles

I'd suggest that the people writing those papers I pointed to don't agree at all

 

[jcx]

CD perceptual dynamic range isn't as limited as some here appear to believe

http://audio.rightmark.org/lukin/dither/

has dithered wav files at 12 and 8 bit to illustrate linearity/perceptual dynamic range extension by dither

and Stuart's paper points out that even more perceptual dynamic range improvement is available with dither designed for use with the standard CD pre-emphasis

Loudness War compression levels have nothing to do with CD audio 16 bit depth or any claimed lack of resolution

 

[Daveis]

Has anyone got comparative data on the differing audio standards between DVD video and HD DVD video either in two channel mode or multi-channel mode?

DVD Video has support 16/48 PCM, Dolby AC3 and DTS. Of those DTS sounds the best to me. Interestingly, the dynamics on DTS are usually very present.

DVD-Audio would have been wonderful since they used the same phsical media as DVD Video. Sadly, the DRM and lack of titles held it back.

HDCD probably would have worked, but was overshadowed by the SACD /DVD Audio war that destroyed both formats.

The next great hope....

From wikipedia...

HD DVD discs support encoding in up to 24-bit/192 kHz for two channels, or up to eight channels of up to 24-bit/96 kHz encoding.[8] For reference, even new big-budget Hollywood films are mastered in only 24-bit/48 kHz, with 16-bit/48 kHz being common for ordinary films.[citation needed]

All HD DVD players are required to decode linear (uncompressed) PCM, Dolby Digital AC-3, Dolby Digital EX, DTS, Dolby Digital Plus and Dolby TrueHD.[9] A secondary soundtrack, if present, can be stored in any of the aforementioned formats, or in one of the HD DVD optional codecs: DTS-HD High Resolution Audio and DTS-HD Master Audio.

For the highest-fidelity audio experience, HD DVD offers content-producers the choice of linear PCM, Dolby TrueHD and DTS-HD Master Audio. Due to the high-bandwidth requirements of linear-PCM, lossless audio on HD DVD movies has thus far been delivered in the lossless format Dolby True-HD.

 

[mrwhy]

There are two things that worry us:
1. The compression during recording.
There is no way to undo this: it is unknown and unrecorded.
2. A specification of what WOULD please us.
Once we had that we COULD do something about it!
We could make and market our own recordings

So to get us started I propose:
Our recording process will pass the following tests:

Intermodulation less than 0.1% rms sum as demonstrated in the following two tests:
A) A microphone exposed to a 1000 Hz sine wave tone together with a tone at 1347 Hz each at the actual peak spl during recording shall produce a recording in which the rms sum power at 347 Hz and all multiples up to 13470 Hz shall together be less than 0.1% of the power at 1000 Hz.
B) As test A but at minus 70 db.

All else is secondary to the above, but we might as well include frequency response flat +/- 2 db from 50 Hz to 25 KHz.

I suspect a cd could readily pass A, but not B.

If current techniques will not pass B, then we need something better than “dithered red labels, etc, etc”

John