High Resolution Audio: Does It Matter?

[sergeauckland]

Sadly, true. I listen mostly to classical which doesn't have this problem. But recently I got Rival Sons' latest album Feral Roots based on a recommendation from a friend. The first few seconds revealed it was so distorted and crunched it was practically unlistenable. So I got curious and ran some tools. Turns out measures a paltry DR5, with massive clipping all over the place. If I were in that band I'd be angry at the recording engineers for screwing up my work and art!
View attachment 23354
The irony is, they have a 96-24 download for an album that has so little dynamic range it could be transferred to 8-bit! I didn't want to unfairly judge it by classical music standards, so I measured Pink Floyd Wish You Were Here for comparison, which measures DR14 with no clipping.

Early CDs, certainly pre 1995, do seem to be much better from the point of view of clipping or heavy compression and limiting. I've mentioned this before, but all my Dire Straits CDs, bought when they first were published, have no clipping and have peak levels several dBs under 0dBFS. Ditto Moody Blues and Santana CDs of a similar vintage. Then TC Electronics invented the Finalizer and it all went to rat-shit.

S

 

[PierreV]

Sadly, true. I listen mostly to classical which doesn't have this problem. But recently I got Rival Sons' latest album Feral Roots based on a recommendation from a friend. The first few seconds revealed it was so distorted and crunched it was practically unlistenable. So I got curious and ran some tools. Turns out measures a paltry DR5, with massive clipping all over the place. If I were in that band I'd be angry at the recording engineers for screwing up my work and art!

The track is "Do Your Worst" - what did you expect
Oh, and it is MQA Studio 96kHz - 24 bits on Tidal

But, seriously, given the number of times I have had to double check nothing was wrong with my speakers lately, I think some distortion is now in vogue and a creative choice

 

[sergeauckland]

The track is "Do Your Worst" - what did you expect
Oh, and it is MQA Studio 96kHz - 24 bits on Tidal

But, seriously, given the number of times I have had to double check nothing was wrong with my speakers lately, I think some distortion is now in vogue and a creative choice

I've seen this especially with the less experienced bands, and/or those who appeal more to a younger and more mobile listener and who want to stand out. They actually ask to be LOUD, and as this is now trivially easy, it gets done. Even if the peak level is kept off 0dBFS, it's still clipped earlier, so it might just as well go to 0dBFS as it's not going to sound any worse.

Here's Rag and Bone Man's Human

 

[MRC01]

A sinc filter is infinite in length. Not sure where you're headed there.
Now, nobody uses that kind of filter, even windowed. There are better ways to design an anti-aliasing/anti-imaging filter that have quite finite length. ...

I'll rephrase my point as a question. If the Whittaker-Shannon formula reconstructs the analog wave theoretically perfectly, and nobody actually uses it, then doesn't that imply that all methods actually being used are imperfect?

This point is in response to andreasmaaan's question in post #486.

 

[dc655321]

then doesn't that imply that all methods actually being used are imperfect

Yes. One might think of it as the "art of compromise", AKA engineering.
Perfection can approached arbitrarily closely, however. But, not without compromise(s)...

 

[j_j]

Sadly, true. I listen mostly to classical which doesn't have this problem. But recently I got Rival Sons' latest album Feral Roots based on a recommendation from a friend. The first few seconds revealed it was so distorted and crunched it was practically unlistenable. So I got curious and ran some tools. Turns out measures a paltry DR5, with massive clipping all over the place. If I were in that band I'd be angry at the recording engineers for screwing up my work and art!

The irony is, they have a 96-24 download for an album that has so little dynamic range it could be transferred to 8-bit! I didn't want to unfairly judge it by classical music standards, so I measured Pink Floyd Wish You Were Here for comparison, which measures DR14 with no clipping.

Yep.

It's bad enough we had to actually build a tool to warn us when we were dealing with material that was utterly screwed up, before we actually processed it.

There are also a bunch of remasters of old analog music, say some of the old acapella rock stuff, wherein they saw it necessary to boost it into clipping.

For acapella music.

My opinion of that very well might violate politeness standards here.

 

[j_j]

I'll rephrase my point as a question. If the Whittaker-Shannon formula reconstructs the analog wave theoretically perfectly, and nobody actually uses it, then doesn't that imply that all methods actually being used are imperfect?

This point is in response to andreasmaaan's question in post #486.

Measurably so, by observing a cutoff frequency that is below fs/2, but that's hardly a major imperfection. Thing is, by looking at the antialiasing filter, it is instantly evident what will be good and what will not be.

One can measure the imperfections, but the Shannon Theorem (note, Nyquest conjecture was proven by Shannon as a side-effect of inventing information theory) still works exactly, and what is "wrong" is evident, and is, in any rational implementation, a slight reduction in bandwidth.

 

[MRC01]

Yep.
There are also a bunch of remasters of old analog music, say some of the old acapella rock stuff, wherein they saw it necessary to boost it into clipping.
For acapella music.
My opinion of that very well might violate politeness standards here.

As a big fan of acoustic music in all its forms including acapella, I whole-heartedly agree! It makes me not want to listen to modern music anymore. I wonder how any self-respecting recording or mastering engineer can do this. His job is preserve music, not to destroy it.

I had a recent conversation with HDTracks, about that album mentioned earlier. To their credit, their first offer was to refund what I paid for the album, even though I didn't ask for a refund. They said they already look for objective flaws but compression, even extreme, is an artistic mastering choice, not an objective flaw. I asked whether they consider digital clipping to be an objective flaw, pointing out that recording had it almost constantly, thousands of instances per song. That's one more thing they can add to their quality checks to keep studios honest and help consumers get better recordings. But...

I suspect they're trapped in the middle between the studios & the customer and they don't have the guts to apply meaningful quality checks that might limit the recordings they resell, hitting their bottom line. Which is sad, given that HDTracks is founded by the Chesky brothers. It's myopic thinking because while it might hurt biz in the short term, it would only weed out bad recordings that don't need "high definition" anyway and in the long run they'd become a more trusted source of true high fidelity recordings.

 

[MRC01]

PS: while DR14 is a useful tool, it's only the overall peak to RMS ratio. I'd like to see another DR rating that gives a more detailed idea of the maximum dynamic range of the recording.

I imagine a metric called "bits actually used for music". For example, slide an N-second window across the entire recording and track the lowest and highest amplitudes of music/data (not dither/noise/zeroes) sustained over the window. Unfortunately, being time-window based can be gamed by using tiny window sizes; N would have to be standardized. The idea is to eliminate the pretense of "high-res" and get truth-in-advertising what is actually being sold. Would anybody actually buy a "high-res" recording if they knew the actual music had less than 8 bits of dynamic range?

 

[wiggum]

PS: while DR14 is a useful tool, it's only the overall peak to RMS ratio. I'd like to see another DR rating that gives a more detailed idea of the maximum dynamic range of the recording.

Why do you need another DR rating? If it is not overall peak to RMS ratio, what other ratio is meaningful? If we compare the peak(placed at 0dBFS) with silence in the track, the DR would be very close to theoretical DR which is 96dB for 16-bit audio, but this value is hardly useful.

 

[dc655321]

PS: while DR14 is a useful tool, it's only the overall peak to RMS ratio. I'd like to see another DR rating that gives a more detailed idea of the maximum dynamic range of the recording.

DR14 is a little more complex than that, but not much.
Had a good thread on that (and other loudness/DR issues) last summer. Was eye-opening for me, at least.

I imagine a metric called "bits actually used for music". For example, slide an N-second window across the entire recording and track the lowest and highest amplitudes of music/data (not dither/noise/zeroes) sustained over the window. Unfortunately, being time-window based can be gamed by using tiny window sizes; N would have to be standardized. The idea is to eliminate the pretense of "high-res" and get truth-in-advertising what is actually being sold. Would anybody actually buy a "high-res" recording if they knew the actual music had less than 8 bits of dynamic range?

Maybe all tracks should come with a histogram of the loudness distribution (perhaps equivalent to your notion)?
A visual check of what you might be injecting into your ears...

 

[MRC01]

If it is not overall peak to RMS ratio, what other ratio is meaningful? If we compare the peak(placed at 0dBFS) with silence in the track, the DR would be very close to theoretical DR which is 96dB for 16-bit audio, but this value is hardly useful.

The problem: Peak/RMS is useful, but it's a rough aggregate metric that can produce the same score for albums having perceptually very different dynamic range.
Yes, digital silence/zeros/dither/noise should be ignored.

...
Maybe all tracks should come with a histogram of the loudness distribution (perhaps equivalent to your notion)?
A visual check of what you might be injecting into your ears...

Nice reference. I'm not surprised JJ came up with it. Yes, something like that. Simply building a histogram with sample count on Y axis with amplitude buckets on the X axis, then expressing the 25/50/75/95 percentiles would probably be sufficient. E.g. 25% of the samples are -20 dB or less, 50% of the samples are -15 dB or less, etc. So 30/20/10/3 would have big natural dynamics, and 12/6/3/0 might be a typical compressed modern pop/rock album.

 

[SIY]

FWIW, Keith Howard has a nice freeware tool on his site to get a DR histogram. The user interface is somewhere south of clunky, but it actually works.

http://www.audiosignal.co.uk/freeware.html Scroll down to "Dynamics Analysis."

 

[Daverz]

I was following that story in the original post with great interest until Bob Stuart showed up. I don't trust anything that man touches.

 

[Sal1950]

I was following that story in the original post with great interest until Bob Stuart showed up. I don't trust anything that man touches.

That's a damn good approach.

 

[watchnerd]

I was following that story in the original post with great interest until Bob Stuart showed up. I don't trust anything that man touches.

That's a damn good approach.

 

[KSTR]

The problem: Peak/RMS is useful, but it's a rough aggregate metric that can produce the same score for albums having perceptually very different dynamic range.

IME, Peak/RMS is not useful at all, it's only a very crude indicator.
Notably on hard-clipped data like the example posted by @sergeauckland it is possible to find and apply an audibly benign allpass that can increase the peak values from clipped snares and bassdrums etc by up to a factor of three(!) while leaving the RMS value untouched. DR goes way up but it doesn't sound any different (actually, the clipping sticks out a bit more in audible terms).
What's needed is a metric based on short-term vs. long term RMS values, sort of a judgement of the RMS envelope, preferably also divided into frequency ranges, say, octave-bands (since most pro loudness-war tools are multiband). So that you get ten RMS envelopes which can be analysed for
shape, along the lines of ADSR curves known from Synthesizers, also looking for correlations between bands. If there is little to no Attack and Decay sections and the Sustain phase is flat over most of its length until it delves down in the Release part then there is little real dynamics no matter what the simple overall (single-band) Peak/RMS DRxy rating might want to tell.

 

[LF78]

As I said in another thread I'm experimenting with Audacity software in "fixing" clipped/low DR recordings, and I'm having some results with following effects applied in sequence.

1) Clip Fix:

2) Limiter:

Here are some before/after results.

Before:

After:

Before:

After:

Before:

After:

 

[solderdude]

The plots look nicer, the signal looks less clipped and more dynamic, the DR meter reports a higher DR number, but the real dynamic range did not change.
What changed is the overall level is lower and the peaks have been 'extrapolated' but differ from what would have been seen before the song was clipped.
It may sound less 'clipped' and those DAC's that have issues handling 0dB could benefit for sure but the overall 'sound' did not change or become more dynamic.

It kind of 'fixes' clipping only.

 

[LF78]

The plots look nicer, the signal looks less clipped and more dynamic, the DR meter reports a higher DR number, but the real dynamic range did not change.

Of course you cannot recreate original data that it's simply not there. But I feel that listening is much easier after this processing, and moreover I don't need to run and decrease the volume when these tracks are in the playlist