Digital Audio Explained (from an engineer's standpoint)

[JP]

I can say things as points, clarifications, or to elucidate my thoughts without having them mentioned previously by you or anyone else. This does not mean I'm putting words in your mouth. If I were, I'd say something deceptively clever, like "you said..."

If someone says 'copy' in the digital realm, that implies the bits are the same, hence bit perfect. The general subject Allen delved in to after the CD comment was digital generations sounding different. If he meant he had a CD produced and it sounded different because the bits were different, we wouldn't be discussing anything. However, he said CD copy.

BTW, I said: "For a copy of a bucket of bits to be degraded, there has to be a difference. Discounting corruption and errors which are easily detected, there is simply no difference in a bit perfect copy."

You said: "That's the theory, isn't it?"

Again, 'there is simply no difference in a bit perfect copy." "That's the theory, isn't it?"

Are you not stating here that you believe my statement above is a theory? Would that not then imply that you don't believe it to be factual and provable?

On the generations test methodology, I did explain it, in that paragraph. I could state it again, but I'd just be saying the same thing. Perhaps you can tell me what, specifically, you don't understand or would like clarified.

When you say "engineers", you mean audio engineers, not actual digital signal engineers, right? You do know that just because someone is a professional or otherwise aware of psychoacoustics, that doesn't make them immune to it? The filtering the brain does on the sound we process and the associated tricks and biases apply to everyone, whether they're a professional or not.

No ABX box.

 

[E-Stat]

I agree, 24/96 is likely the sweet spot. Very dynamic, clean and virtually no noise.

And I love the ability to download selections at will into my digital library. I'm an old boomer but get the instant gratification allure that iTunes offers with its lossy content. Most often, I order CDs from Amazon, wait a week, rip them and file the jewel case away most likely never to be used again.

The added storage "cost" of 24/96 is virtually zilch today. My entire 800 odd album library consumes only 170 GB. I have four additional copies stored on inexpensive USB external hard drives.

I just don't get why folks are so adamant against getting better quality - even if they don't immediately perceive it. Go figure.

 

[E-Stat]

I can say things as points, clarifications, or to elucidate my thoughts without having them mentioned previously by you or anyone else.

Some folks might have thought that your comments in post #41 actually reflected what someone actually said!

If someone says 'copy' in the digital realm, that implies the bits are the same, hence bit perfect.

We are quite aware of your assumptions.

The general subject Allen delved in to after the CD comment was digital generations sounding different. If he meant he had a CD produced and it sounded different because the bits were different, we wouldn't be discussing anything.

Please reference those comments with a time stamp in the video so that we can put your comments into perspective.

You said: "That's the theory, isn't it?"

I really wish folks would quote what I actually said so that my comments could be placed in perspective.

Are you not stating here that you believe my statement above is a theory? Would that not then imply that you don't believe it to be factual and provable?

Without any context of copying procedure, media used, evaluation involved, your comments are merely theoretical and have little meaning.

When you say "engineers", you mean audio engineers, not actual digital signal engineers, right?

Those who work with recording digital music every day.

No ABX box.

Please explain your methodology for downsampling and comparison. Did you miss my question about your "10,000 generation" comment and how "we" evaluated it? Please detail that process.

 

[Reynolds531]

The formula that I know and understand is that Improvement in Sound quality is inversely proportional to the to the number of graphs and equations used to convince me.

Using this relationship indicates that the sound quality improvement of SACD over Redbook approaches zero.

 

[jberger]

If all amplifiers sound the same, then why don't you find companies that produce $300 amplifiers tout their success against $30,000 amplifiers? Your guess is as good as mine.

Isn't that what Bob Carver did with his Magnetic Field Amps, comparing them to Conrad Johnson (and others)?

 

[JP]

I'm not going to take them time to quote you or suffer through that video again to reference timestamps for you. The thread isn't that long, people can go back and look if they want to.

Since you feel it's my assumption a digital copy would be bit perfect, why don't you either A) tell us why it's not or B) outline a copy scenario so we know what you're talking about.

So, yes on what you meant by engineer?

Generational test:

1) PCM file of the audio to be tested. I don't remember what song it was.
2) Make 10000 generational copies of the file. B is a copy of A, C is a copy of B, D is a copy of C... 10000 times.
3) Generate and compare MD5 checksums of all the files.
4) Use ABX software (don't remember which) to ABX file 1 and file 10000.
5) Use ABX software to ABX randomly selected files from the set. I believe we did 5 sets of these.

Soundcard was an RME. No downsampling. If you downsampled it wouldn't be a copy, it'd be a different set of bits.

Though on that note, you really should take a 24/96 song and downsample it to 16/44.1 and ABX it. Post the files and your results so others can play. You may be surprised at the results.

 

[E-Stat]

Since you feel it's my assumption a digital copy would be bit perfect...

That would be your assumption. Perhaps you've forgotten comments in post #46:

"Using my own words, a copy, if it's truly a copy, would be bit perfect."

Have you changed your mind since then?

Generational test:

1) PCM file of the audio to be tested. I don't remember what song it was.
2) Make 10000 generational copies of the file. B is a copy of A, C is a copy of B, D is a copy of C... 10000 times.
3) Generate and compare MD5 checksums of all the files.
4) Use ABX software (don't remember which) to ABX file 1 and file 10000.
5) Use ABX software to ABX randomly selected files from the set. I believe we did 5 sets of these.

Soundcard was an RME. No downsampling. If you downsampled it wouldn't be a copy, it'd be a different set of bits.

Copied to what? Using what software? The context is generational copies of CD-ROM media. Is that what the test that "we" did use? Who is "we"?

Though on that note, you really should take a 24/96 song and downsample it to 16/44.1 and ABX it. Post the files and your results so others can play. You may be surprised at the results.

Perhaps you missed an earlier post where I mentioned that I have about half a dozen recordings in multiple media. The 24/96 has the bass response and noise level of the CD and the high frequency extension of the LP. I merely state the obvious.

 

[E-Stat]

Isn't that what Bob Carver did with his Magnetic Field Amps, comparing them to Conrad Johnson (and others)?

It was only against a C-J amp. That's a great story. In The August 2008 issue of TAS, he was asked about that test. His response?

"I should have said, I can make my amplifier sound close...

Well, I have a secret. I cheated. I practiced a lot before I started! There's a lot going on behind the curtain."

And afterwards, he designed and sold his own tube amps. Go figure!

 

[JP]

Look above - I said a digital copy is implied to be bit perfect and you said "we are quite aware of your assumptions." I'm not chaining my mind, I'm saying if you feel that's an assumption then define it as you see it so we're talking about the same thing. Unless you're being argumentative for the sake of being argumentative.

Hard drive. It was Linux, and most likely a shell script, so cp. "We" was a previous company. Who specifically isn't relevant.

I hold the same is true for CD. Setup reasonable controls - same burner, same batch of media, same playback, etc. If the bits are the same then it'll sound the same. If the bits aren't the same then it's not an exact copy. Any half competent shop can analyze that, and if there are excessive soft or hard errors. "We" used to all the time. All the issues we saw were due to bad batches of burnable media, failing burners, or mastering mistakes (not as in the PCM, but the disc itself). Never had an issue with the stuff done off glass unless it was a mastering mistake.

And yes, you said you've recordings on multiple media, but you didn't state that you down sampled them yourself. You're not assuming they all use the exact same master without changes? It's very common for audio engineers to make different masters for different media. I'll tell you what - tell me what song you want that's available in high-res download, and I'll downsample it for you. What do you have to loose?

 

[E-Stat]

Hard drive. It was Linux, and most likely a shell script, so cp. "We" was a previous company. Who specifically isn't relevant.

Then you're comparing apples to pomegranates. I really didn't think that anyone would actually make 10,000 generational copies via CD media. Thanks for confirming that. Which is what Allen was referring to.

What do you have to loose?

What do I have to loose?

Here is music I have on CD, 24/96 and in many cases, LP:

Madonna - True Blue
Goldfinger Soundtrack
Vince Guaraldi - A Charlie Brown Christmas
Chicago II
Heart - Dreamboat Annie
Stan Getz - Getz/Gilberto
Moody Blues - Days of Future Passed
John Williams - Jurassic Park

 

[RobInONT]

But audible. Does the difference exist? Yes. Does it matter to you?

Did you notice the frequency that this single cycle of ringing occurs? It occurs close to Nyquist (22050 hz in the image you've provided). It's not audible except maybe to your dog. Actually, even your dog isn't going to hear a single cycle of 22000 hz. Like I said before, the "problems" with the redbook standard that you like to keep dredging up have been dealt with long ago by oversampling and digital filtering.

 

[JP]

It's not, and that's the point. Be it hard drive, USB, CD, what have you, bits are bits. Allen made a specific comment about copying a CD (not production, he just said copying) and then continued further off the road and spoke in general about digital generations.

So, take a stance. Do you believe in digital generational degradation for files? For CDs? For either if the bits are the same?

Let's do Getz/Gilberto. Who's 24/96 copy of it do you have? Any track in particular?

If anyone has interest in a broader test, let me know as we'll have to find something that is royalty free to use.

 

[E-Stat]

Did you notice the frequency that this single cycle of ringing occurs? It occurs close to Nyquist (22050 hz in the image you've provided). It's not audible except maybe to your dog.

That's using the "listen" filter that rolls of HF which is not commonly used in CD players. The vast majority of CDPs exhibit multiple cycles of pre-ringing when attempting to run flat to 20 khz.

Like I said before, the "problems" with the redbook standard that you like to keep dredging up have been dealt with long ago by oversampling and digital filtering.

Given that 99% of CD players don't use this compromise - yes Virginia we no longer have flat response to 20 khz, then you still don't get it. Higher sampling rates don't require such compromises.

By all means, dumb down the experience however you please!

 

[shelly_d]

I have read through this entire post but since I am at work I cannot access the interview. I will address 2 narrow points, generational degradation and the conclusions of the Meyer/Moran testing.

The reproduction of a CD master to CD copies should be done digitally. If it isn't then the entire production team should be dismissed. Once into production, the digital copy should be, as JJones3 stated, bit perfect. If not, then it is dropping bits or adding them due to noise or some other fault in the equipment. Either is a glaring defect that needs to be fixed. It is very easy to run checksums on digital information and then compare them. It will find differences in files with beyond human accuracy and repeatability. If subsequent generations of CD's have errors, it is because of faulty equipment that needs repair or replacement. It really is that simple.


The Meyer/Moran test very clearly observed a difference in the source material for the high res formats and the CD world. They clearly identified where the difference came from, and that is the care and attention to detail observed in the mastering process. I would point out that a given release by a given artist is done differently for CD, vinyl, and hi res digital. They also amply demonstrated that "brick wall degradation" was not readily heard by their panel of listeners. While no test of this nature is conclusive, and to be considered a valid conclusion it would need to be duplicated with similar results over a statistically valid sample size, it is nevertheless quite important and revealing.

 

[E-Stat]

Allen made a specific comment about copying a CD (not production, he just said copying) and then continued further off the road and spoke in general about digital generations.

Exactly. When you make a copy of a CD, you get another CD. Hard disks don't apply.

So, take a stance. Do you believe in digital generational degradation for files?

I'll continue to observe my experience and that of others. I've copied files using spinning rust and haven't noticed any degradation. When generating multiple generational copies using shiny aluminum removable media, however, yes I have.

Let's do Getz/Gilberto. Who's 24/96 copy of it do you have?

That would be the version available on HD Tracks. Girl From Ipanema would be fine.

 

[shelly_d]

That's using the "listen" filter that rolls of HF which is not commonly used in CD players. The vast majority of CDPs exhibit multiple cycles of pre-ringing when attempting to run flat to 20 khz.


Given that 99% of CD players don't use this compromise - yes Virginia we no longer have flat response to 20 khz, then you still don't get it. Higher sampling rates don't require such compromises.

By all means, dumb down the experience however you please!

Please LOOK at the test setup the study used. They ran an analog signal through both an A/D and then a D/A conversion and compared that to the input analog signal. That really is the toughest condition for the CD standard for a valid comparison. It was the same A/D D/A conversions that you are talking about. It had all the same impairments you are talking about. The listeners could not hear it.

Thus conclusions that one might draw from this are:

1. There is no significant degradation to the signal in the D/A A/D conversion done,

or

2. The test subjects were insufficiently critical in their listening to hear the differences,

or

3. Only people with extraordinary hearing can hear the degradation

or

4. The degradation really is a thing of the past.

 

[JP]

Okay, so you don't believe in generational degradation for digital in general, just CD? How was it copied? Did you verify it was in fact a good copy - bit perfect?

I'll have to buy it. I've the AP SACD and LP. PM me your email address so I can send you a dropbox link when I get it done.

 

[dumptruck]

Just use 10 second samples if you want to post a test - you'd have to be crazy to claim that's not fair use when there's a specific purpose for it. I'm not really interested; I've already failed that test, as well as MP3s 256kbps and up, once I started trying actual ABX for such things, and I ain't ashamed to say so. If I need an ego boost, I'll just think about how I can perform far above average on pitch perception and distortion hearing tests . It was kind of thrilling, in a way, having my certainty about hearing differences evaporate in a matter of seconds. No amount of non-blind A/B gets you that feeling.

If you do post a test and a bunch of people say they passed it, I'll perk up, and give it go too. After seeing many, many proposals for such tests and even starting a couple of threads proposing them with instructions, I would say don't get your hopes up.

There's also the issue of getting it done. With a studio setup and pro audio software, it's a piece of cake: if you don't already have an automated plugin, there's a still going to be a simple enough way to use A/B functions for ABX. Otherwise, it's a bit of a pain and has very setup-specific concerns. Redbook vs lower is a little easier, since you can pretty much use the foobar ABX for anyone with a computer source (still potentially useable for higher quality, but OS/driver issues come into play). Whatever you do, don't forget statistics. 8/10 correct doesn't mean anything, unless you get it over and over, but then it's not /10 anymore.

Here's a related rather hopeless thread I made recently (new year's eve afternoon, apparently) http://audiokarma.org/forums/showthread.php?t=565342

 

[E-Stat]

Please LOOK at the test setup the study used.

It was pretty crappy, wasn't it? Do you use a $300 player?

It was the same A/D D/A conversions that you are talking about. It had all the same impairments you are talking about. The listeners could not hear it.

Apparently, their panel could not hear what the Ayre engineers could in developing the alternative filter position. Ok.

Thus conclusions that one might draw from this are...

Or...

5. Running arbitrary SACD content through mediocre gear using ABX boxes renders the test an absolute joke.

Did that BAS test have any effect on the industry recording process or production of any audio equipment? Just lots of smiles.

 

[E-Stat]

Okay, so you don't believe in generational degradation for digital in general, just CD? How was it copied? Did you verify it was in fact a good copy - bit perfect?

The issue has always been about CD copies. I used EAC and Roxio. Now use dbPoweramp for ripping.

I'll have to buy it. I've the AP SACD and LP. PM me your email address so I can send you a dropbox link when I get it done.

Will do.