Dynamic Range, Audacity graphs, and High Res Question

[esldude]

I did not know Tusk was done on a Soundstream. I knew Soundstream was in use, but didn't know Fleetwood Mac had utilized it at the time.

 

I loved Rumours. It had a nice quality that set a certain mood while listening to it. It never was a high quality recording in my opinion. Sounded more muddled than average though the resulting music worked in spite of that. Tusk was clearly a better recording even on the LP in my opinion back at that time though I didn't know it was digital. On the other hand I simply didn't like the music on it as much.

 

I do remember in the early 80's while many didn't like digitally recorded LP's I never heard a Soundstream recording I didn't think pretty highly of. Didn't always care for the music, but I never agreed with complaints about its basic sound quality.

 

[bleedink]

I was surprised as well. I always assumed it was an analog recording based on the time it was recorded and that FM always seemed an 'analog' sounding band until about mid 80s. Still trying to digest some of this excellent seminar on dynamics and such. It's always great to have some idea what you guys are talking about from time to time!

 

[Miska]

is that those low level signals will be significantly lower in resolution than those near the top of the dynamic range

 

Now, this begins to depend a lot on the encoding algorithm, namely dither and noise shaping used. With good algorithm and high enough sampling rate it behaves much like analog.

 

All modern A/D and D/A converter chips use only 1-6 bits internally at the conversion stage. A/D's typically running the conversion stage at 5.6/6.1 MHz. All the rest is just DSP algorithms. (IIRC, MH uses AKM converters that have been traditionally low bit depth ones.)

 

So I'll just repeat my earlier statement, that resolution is product of sampling rate, bit depth and algorithm. None of those matter alone.

 

The same recording on vinyl will not suffer any of this.

 

On vinyl the limiting factor is when those harmonics end up being masked by the noise floor.

 

Same happens with good digital encoding.

 

 

[PeterSt]

Hi there Barry,

 

[in the mean time there's Miska's post, but I think mine may fit to that]

 

Thank you for responding. I think it is great.

Well, I referred to our earlier little talk about this, and it is right there were it stopped for me. I just didn't get it, and still don't;

 

While one can, in theory record a very low level signal on CD, what often is not taken into account, yet is quite plainly audible, is that those low level signals will be significantly lower in resolution than those near the top of the dynamic range.

 

Let me say first that I can make mistakes, or not think about everything and all, or better put : I would like to find out how this works (or how it works in your perception; also good). So, I will not disagree with you, but I just don't understand, ok ?

Now :

 

Suppose there is a bottom end of the signal. Of course this is infinitely low levelled, but let's say that drummer will not move the brushes so softly that it will be beyond the environmental noise, molecule noise or other electrical noise. I think I also should exclude A/D or later D/A noise (but this is dangerous, because possibly the subject).

Let's say this is the 30dB we talked about earlier in this thread. So, 1dB above that is assumed audible to everything (ears, microphones, A/D and later D/A). Thus 31dB is audible.

 

What you seem to say is that it is better to crank up levels to e.g. 67dB (48dB more just for conviently talking about it), so 6 bits of more resolution are available to capture that 1dB of sound above the noise. No, this is not what you are saying, but this is what it would come down to. Or ?

That the subject now becomes "noise" as such, I can't help. But I really think that the 30dB of environmental noise is amplified with 48 dB too. Btw, "environmental noise" is not to be audience noise and the like; so think molecule.

 

The sound is still 1dB above the noise, and that the noise now plays at 66dB is not a problem. All is relative, and all depends on how much the volume is open later (playback). Maybe we now will be able to hear the inherent noise while without this additional 48dB level during the recording it would not, but I think this can be excluded from the equation (only when Mahler is played way too loud for its upcoming very loud passages it would play a role, but regarding the as upcoming breaking windows, we won't. So, the noise issue regarding this is not a subject (again, I think).

 

What you now say, as it seems, is that the now available 6 more bits allow for a better resolution of that 1dB music level, while it seems to me that already all the (24) bits are available right from the start for it. Of course, we could say that the lowest 30dB - 5 bits comprising noise are not utilized, while the additional 6 bits now can take care of that 1dB of music level.

 

Regarding the latter, it seems obvious that 6 bits for 1dB is better than 1 bit, but something ain't right;

Remember, I sure believe what you perceive.

 

Now, we also assume (this is to make the subject somewhat more interesting) that no dither is applied. You didn't talk about that to backup what you perceive, so you don't. :-)

 

Also assuming that indeed it will be so that the 30dB of noise is linearly (48dB) amplified with the signal (which I am not sure about, but for my own fun I take this for granted now), there is this 1dB of music and 66dB of noise. Ok ...

 

To me this looks like 10 bits being available for dither.

... Dither which is already there in the most perfect way, because it would be totally random; It is analoguely amplified molecule noise ...

 

Did I just solve my own quest ?

 

It tells me additionally that it well may be so that the 6 bits of amplification can have nicely captured all that 48dB of amplified music. No matter the noise is amplified with it. This is how dither works, although I put its working upside down. Of course, the original 1dB of music (now being 49dB) would be full of noise, but this is not perceived necessarily; it is playing at 144 - 67 = -77dBFS for its top end only. But the resolution would be there ...

 

Ok, it is clear that I made up the whole thing during this writing. But can it be debunked ? may dither have its working like this ? If so it is the explanation. Or it can be.

 

In the mean time not to forget (or otherwise my subject isn't clear from the start) : lower level sounds do not need the higher level bits. They are just not utilized. 1dB of music level officially needs 1 bit only (even 6dB can do it with that - no, *will* (!) do it with that).

And so I think it really needs a story like mine above.

 

Peter

 

[PeterSt]

PS: I can recommend Turtle Records for the highest utilized dynamic range, measured by transient steepness (just redbook). For their recordings they used a special reel to reel machine of which I didn't quite get the details but to me it looked like a 60's device (which was still in use a few years ago).

 

It was a Nagra.

 

 

 

[goldsdad]

"I'm not sure that anyone has really provided me the examples I was looking for"

 

In my opinion, the problem with fulfilling your request is that although some information can be interpreted from Audacity graphs, such as brick-walled dynamics or suspected upsampling, there are things which can not, such as scale and solidity of soundstage and realism of timbre of instruments.

 

More importantly, I believe, we each have our unique preferences and weights of significance, whether consciously or not, for different aspects of sound which vary between alternative masterings of a recording, such as tonal balance, macro dynamics, micro dynamics and temporal resolution.

 

A mastering with large dynamic range may lack minuscule details and, to a particular listener, sound uninteresting relative to a dynamically crushed mastering which is otherwise highly transparent to the original recording. Another listener may be more moved by the more dynamic version. One mastering may be too bass heavy for a listener's taste while an alternative mastering is too bright for another listener's taste. And so on.

 

Of course, alternative mixes of a recording add further complexity to the picture.

 

 

 

[bdiament]

Hi Miska and Peter,

 

As I stated a number of times in my post, I am speaking specifically of 16-bit. To be clear, I am not speaking in terms of theory but in terms of real world practice, every day.

 

You can change the A-D converter, the algorithm, the sample rate, add dither if you like, in my experience, it doesn't matter. To my ears (note, I'm speaking for myself here), outside of the capability for steeper, faster dynamic gradations at the low end, 16-bit is an order or magnitude behind decent vinyl when it comes to dynamics, particularly microdynamics that lie within the larger gradations. And it is thus, an order of magnitude behind decent vinyl when it comes to harmonic presentation and spatial cues.

 

Same with vinyl; I'm talking listening experience and not theory. That harmonics are there to be heard quite clearly. The noise floor is not a problem. (Of course, with particularly bad pressings or particularly bad turntables, the point becomes moot. I'm talking about a decent or better setup with a decent or better pressing.) Decades of listening to vinyl and getting used to hearing the harmonics that were on the master tapes I heard at the studio, resulted in quite a shock when I first heard "perfect sound forever" was actually "quite bad sound, right now". ;-}

 

I'm not talking about "harmonics masked by the noise floor"; unless the system is a very bad one, the noise floor will be well below the harmonics and spatial cues that get lost quite fast with even the very best 16-bit.

 

As always, just my perspective of course. I understand you may hear it differently.

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

 

 

 

[Jud]

Some folks up quite early (late?) on a Saturday to share their thoughts here. :-)

 

If I might try to paraphrase my understanding (very possibly a misunderstanding) of what is being discussed -

 

It seems to me Barry may be lamenting the fact that 6 bits leaves only 2-to-the-6th or 64 possible gradations in loudness levels. (Is that possibly correct, or am I way off here?) If one thinks about color levels on a video monitor, 64 possible gradations won't look at all realistic. So although both LPs and CDs might have, say, 48dB of room between the loudest sound and the noise floor (or the LP might have somewhat less), *within* that space the LP will be capable of nearly infinitely fine variation (limited only by the capabilities of the devices in the recording and playback chains and how well or poorly done the recording was), while the CD must present us with the sort of poor imitation you would see of a photograph rendered on a monitor with only 64 possible color levels.

 

[bdiament]

Hi Jud,

 

If I may, no, I'm not lamenting anything about bits or any theoretical potential they might have. I'm addressing the audible consequences of 16-bit vs. vinyl encoding.

 

Also, I'm not simply speaking of a greater number of gradations with vinyl over 16-bit. I'm speaking about what one actually hears when listening to (rather than theorizing about) properly done vinyl vs. properly done CD.

 

With vinyl, harmonic information and spatial cues don't diminish as recorded level diminishes, as they do with even the finest 16-bit in my experience.

 

This was the point of my first post in this thread:

The idea of CD having greater dynamic range than vinyl is a theoretical concept that is not born out in practice - at least not for my ears.

 

To be clear, I'm not talking about signal to noise ratio (as I believe many, who use the term "dynamic range" in fact are). I'm not even talking about loudness gradations. I'm talking about degraded sonics at lower levels, which to my mind is an indication that the real dynamic range of 16-bit does not reach downward to anything like what the signal to noise ratio suggests.

(I think that last sentence sums up what I've been trying to say for the past three or four posts.)

 

One can throw any sort of dither (and/or noise shaping) at the problem they'd like. A sound that is ~30 dB down from max peak is encoded with ~11 bits of resolution. No amount of dither is going to change that, hide that or (again, to my ears) in any way make that better. Anyone interested in 11-bit digital?

 

Play the same thing on a decent vinyl rig and the harmonics and spatial information are still there. They haven't been coarsened or eradicated.

 

The only way around this with digital is to add more bits. That's why we have 24-bits today.

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

 

[Jud]

Thanks, Barry. Even though your last comment wasn't saying things very much differently, perhaps it was just the accumulation that's made it clearer to me now. I think I'll go listen to some LPs (after I get done mowing the lawn :-).

 

[Mark Powell]

A British magazine, HiFi News and Record Review, has for the last few issues been testing all the 'Hi Res' music that they review. The test suite they use is precisely the same as is used by many High End manufacturers around the world and is operated by the guy who designs it.

 

To date approximately one third have been found to be upsampled 44.1 or 48 files. The tests show clearly that it is the larger vendors who are proportionately the most guilty.

 

It is noticeable that none of the vendors, including some of the most well known, have disputed the test results. But they have written in and made lots of ludicrous 'excuses'.

 

They are as free to buy the non-disputed test equipment as easily as anyone else so pretending they didn't know does not hold water. I could name names, and give you the exact numbers so far (they cannot sue me for telling the truth) but I might be seen, by you, I don't care about the vendors, as picking them out.

 

[PeterSt]

Barry, sadly and somehow, the noise (floor) seems to be your subject, while it wasn't mine nor anyone's specifically. And I so much tried to tell it.

Also I don't know why the 16 bit is so important, while as far as I can tell it wasn't a real subject in here at all.

 

It should have been so that you were called into the thread because of my reference to the similar subject of the recording level in another thread many months back.

I even indirectly accuse "you" (recording engineers) of compressing at all times, although I explained with all my power it is a necessary thing.

 

What I brought up is not exactly the thread's subject, but it *is* about dynamic range stuff. Besides that, this (thread) is about 24 bit recordings - or malversations if you like.

 

Why change the subject ?

But it's okay, because I won't tell you what to do of course.

Anyway, it was you (again) telling about the too low level of recording not being a good thing. If this is empirical finding only (no theories at hand), again, fine.

But maybe admit it ?

The only thing I try is explaining what you sure perceive. Now don't tell me that this is with 16 bit recordings ...

 

Of course, with a good 24-bit recording, none of the above applies.

 

Yes, I heard that. But the context where you said that was slightly different I think.

 

Kind regards,

Peter

 

[bdiament]

Hi Peter,

 

I'm sorry if I wasn't clear. I thought I was being quite clear.

No subject has been changed.

And it isn't noise I'm talking about.

It is the confusion between signal to noise ratio and real dynamic range I'm talking about.

 

If you read it again, I hope you'll see my first post in this thread was a direct response to something the OP said:

 

"...However theoretically at least the LP should have less of a dynamic range right?..."

 

That is precisely what my posts in this thread have addressed.

It is an area where I've seen a lot of misunderstanding and general confusion.

I believe this comes from looking at only a part of the theory and not the whole picture.

 

Brings to mind the words of Alfred Korzibski:

The map is not the territory.

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

 

[Miska]

It seems to me Barry may be lamenting the fact that 6 bits leaves only 2-to-the-6th or 64 possible gradations in loudness levels.

 

Yes, in the transfer function for the almost infinitely short period of time. Now, if you take two values and flip between those fast enough with suitable statistical distribution and calculate average, you can produce any intermediate value between those two.

 

This is how modern A/D and D/A converters work. Because more you have bits, more you have real world manufacturing tolerance problems between the corresponding levels. With one bit conversion, you don't have that problem practically at all.

 

For example, with one bit precision and four samples, you can already have quite many output values (simplified version):

 

[0 + 0 + 0 + 0] / 4 = 0

[0 + 0 + 1 + 0] / 4 = 0.25

[0 + 1 + 0 + 1] / 4 = 0.5

[1 + 1 + 1 + 0] / 4 = 0.75

[1 + 1 + 1 + 1] / 4 = 1.0

 

How those bits are ordered is practically noise shaping.

 

 

[Miska]

real dynamic range of 16-bit does not reach downward to anything like what the signal to noise ratio suggests

 

How can you be happy with Metric Halo and it's probably 3-bit conversion stage?

 

Just look how any modern converter works. It's not that straightforward with number of bits you think it is.

 

One can throw any sort of dither (and/or noise shaping) at the problem they'd like. A sound that is ~30 dB down from max peak is encoded with ~11 bits of resolution. No amount of dither is going to change that, hide that or (again, to my ears) in any way make that better. Anyone interested in 11-bit digital?

 

I produce test files encoded with four bits, six and eight bits all the time and those sound just fine. That's something I commonly use for testing the algorithms.

 

I can encode -30 dB signal even using 1-bit. No problem. DSD is doing it all the time...

 

 

[PeterSt]

[barry, I read your post over and over again; of course, if you now are saying that you just responded to what you just told in your last post, what can I say :-) ... I added this little phrase after typing my post below]

 

The idea of CD having greater dynamic range than vinyl is a theoretical concept that is not born out in practice - at least not for my ears.

 

I don't think this would be something I would come up with (that the dynamic range of CD is higher than vinyl), but since it seems to be a subject anyway ... my ears tell me different. Very very different.

But possibly it is needed to read back earlier in the thread how this can be possible in *my* specific situation.

 

But it is easily backed by theeories which are practice, and the theories are that within one sample period the close to 75% of the dynamic range (minus to plus or the other way around !) can be covered for. I mean, in real life CD albums. Most of them lay within 50% of the full range for many samples.

 

It may come to you as unbelievable, wrong, or a wrong workout, but it is just a fact. Maybe it is not easy to find the software which checks for this, or it may be difficult to check it otherwise, but my software does it (as a protection means against my own wrong programming).

 

Of course if this happens from one sample to the other, and we talk about 16/44.1 indeed, this is because the sample resolution is not high enough. But it really doesn't matter, because with e.g. 24/176.4 the rise/fall will be within the same time period.

 

Sidenote : we may think that this is caused by unfiltered means (and a 20KHz frequency could already do this with sufficient amplitude), but I am taking the measure behind the filtering.

 

What does this say ? well, that this is present in CD albums is not so much important (but it was the most surprising to me at the time), but the importance is : I don't see vinyl do this.

(and no, I don't see vinyl do 16Hz either, but this is a kind of another subject ... ehh ...).

 

Before the debate starts how wrong this might be (A/D anomalies for example), it is moot before the subject starts because those dynamics just *are* in there (and when false, they should create piles of harmonic distortion). I can't help that.

The only thing I can help is that in my situation this is not smeared into something you would perceive as less dynamic than vinyl (about I think my first post in this thread).

 

All 'n all I think we must be the most careful about binding empirically found "theories" - but better call it judgements, which are just derived from playback systems.

And no, I did not say my system is better than yours. Only that there is no way I perceive more harmonics or anything than vinyl. Vinyl flaws totally on exactly this.

In my system.

And with my ears of course !

 

Peter

 

 

[bleedink]

"Some folks up quite early (late?) on a Saturday to share their thoughts here. :-)"

 

And I just woke up! Afraid the last few comments in the thread have completely lost me (though I am getting better at figuring out what you guys are talking about!!! :-)