A listener's caveat

Hi all,

I'm not sure if this belongs in Hardware or in Music. I'll make my guess and place it in Hardware but if the Gorts feel differently, I can also see it as relevant in Music.

***

I see a lot of posts from folks who seem to draw sweeping –and erroneous- conclusions based on certain characteristics of some engineering/mastering tools and that has prompted this thread.

Some examples are with regard to analog vs. digital EQ, analog vs. digital compression/limiting, tubes vs. transistors, high sampling rates vs. 44.1k sampling, etc..

Starting with EQ, some folks seem to believe there are certain inherent sounds to analog EQ vs. digital EQ. At one time, I might have agreed. Twenty years ago, all the digital EQ I heard was “hard”, “gritty”, etc.. For this reason, when I mastered a CD with EQ, I would apply it in the analog domain, prior to digitizing. But even in the analog domain, every equalizer sounds very different from every other equalizer. Some (even some expensive pro models) are to my ears, quite awful and superimpose an editorial coloration on every signal passed through them. A few are relatively transparent.

Same with digital EQ today. Some are quite awful and superimpose an editorial coloration on every signal passed through them. A few are relatively transparent. Among these today, there are some that are more transparent to my ears, than even the very best analog equalizers in my experience.

The point? It isn’t analog vs. digital EQ, it is WHICH analog EQ vs. WHICH digital EQ.
To believe one is inherently better than the other is to ignore what some designers have achieved.

Exactly the same thing can be said for digital compression/limiting, tubes vs. transistors, high sampling rates vs. 44.1k sampling, etc.. It isn’t the parts, it is how the parts are used. In each category, it is easy to find examples of excellence and examples of the opposite.

Some folks ask about the chips used in certain DACs (digital to analog converters). In my experience, it isn’t the chip, it is the overall implementation, the analog stages surrounding it, the clocking, the filtering, the phase integrity, etc. A designer can use the “best” chip and still create a bad sounding converter. Another designer might use an “ordinary” chip and come up with a much better sounding converter.

Some folks talk about doing digital encoding at high sample rates, whether 2x like 88.2k or 96k or now, 4x like 176.4k and 192k. Is 192k better than 44.1k? In a few converters, I would say absolutely and it represents a new level of possibilities in recording, removing artifacts I’ve heard with digital recording for more than a quarter century. In other converters, it doesn’t sound as good as the same converters running at 44.1k. Again, it isn’t the characteristic, it is the overall implementation.

There is one popular pro converter that simply does not do a very good job. It is spec’d for 192k and is, after all, a “professional” (add reverb after that word) device. But it is not phase coherent and it does not utilize a monolithic clock to control all processes. It can lead some pros to publish “white papers” (more reverb) on why 192k is simply market-speak.

I often see articles and posts referring to “integer” sample rate conversion being “better” because the math is easier. (Integer conversion implies sample rates that are even multiples of one another, for example, 88.2k being exactly double 44.1k.) Depending on the sample rate conversion algorithm, this might be true but is not by any means necessarily true. A older or lesser SRC algorithm might indeed have an easier time of it with integer conversion. In my experience, however, the best SRC algorithms will perform better regardless of whether the conversion is integer or not. In other words, algorithm A might create a better 44.1k result if the original is at 88.2k and not so good a result if the original is at 96k. Algorithm B on the other hand, can do a better job at either rate, even the more “difficult” 96k than algorithm A can at the “easier” 88.2k. Once again, we see it isn’t the rate, it is the implementation, the overall design of the algorithm that will determine the outcome.

So when we hear a recording or mastering session utilized only analog EQ or only digital EQ or was done at 192k, it might be superb or it might sound like chalk on a blackboard. The terms “digital EQ” or “analog EQ” or “192k” or such-and-such converter chips mean absolutely nothing in themselves. There is no way to determine whether these have a positive or negative influence on the finished product without a lot more information.

And that just covers the hardware/software. At least as large an influence comes from the engineer using the hardware/software. It is not difficult to use the best designed hardware/software and come up with a bad sounding result. The evidence is scattered in all of our collections. I’ve heard good engineers get better results from less than top shelf gear than many other engineers can get in a roomful of state of the art.

So be careful of shoot-from-the-hip posts as well as published reviews declaring something “good” (or “bad”) based on a single isolated characteristic that in the end tells us nothing. There is simply too much variation within each category and within each user to reach a useful conclusion without a lot more information.

Just my perspective.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Barry could you recommend an inexpensive digital eq?

 

Thanks for the insights Barry

 

different levels of inexpensive..... you might want to throw out a dollar amount.

 

I'd like to hear recommendations on Digital EQ regardless of cost, what do you consider the best. Thanks for your insights also.

 

Hi Tullman,

Depends upon the platform you're using and what the budget is.

Some interfaces come with very nice algorithms in their included software but I'm guessing you mean a separate plug-in.

While I use different platforms, I do all my audio on my Mac and as far as separate EQ plug-ins, I haven't heard a match for Metric Halo's (80-bit!) ChannelStrip yet. (It also includes a compressor and a gate.) ChannelStrip comes in two versions, the full version costing $345 and a GarageBand version that sells for $89.

If you're on a Mac and one of these fits the budget, I'd suggest trying a demo.

***
In keeping with the theme of this thread, I would not assume ChannelStrip sounds as good as it does merely because it does 80-bit math. That is merely one of the decisions the designer made in creating it.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Hi Russ,

Its just my take on it of course, not Universal Truth. ;-}

Aside from ChannelStrip (the plug-in), there is a version of this that is part of the software included with Metric Halo's interfaces (the dsp versions), called MIOStrip (as well as MIOEQ6 or MIOEQ12, 6 and 12 band parametrics without the compressor and gate included in MIOStrip or ChannelStrip). I need to run some more head-to-head tests but MIOEQ/MIOStrip might be just a tad more transparent.

For my ears, as of April 2009, any one of those is my first choice when I want the cleanest, most transparent EQ possible, regardless of price or design.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Barry,
Thanks for your insight, thoughts and your experience with these items.
I agree and have often wondered, though I usually don't chime in, why some think that if one product utilizes the latest chip or whatever think that should be the sole issue which sounds better.

I used to think that any CD player that used op-amps, rather than a discrete analog section, had to be the inferior product. It turns out some highly regarded CD players utilize op-amps, while other do not. Both seem to able to sound good, and some don't.

I could come up with other analogies but I think most will get my point. Some time back I read another put it like this; Give two chefs the same ingredients and you'll get two results. Implementation and individual design accounts for much more of the overall sound of a product than the individual parts.

Regards
Terry

 

This post was needed! Thanks, Barry!

 

Barry,
I agree with most of your thoughts here. IMHO, specifications for any component (amps, DAC’s, speakers, etc.) are important to know because they provide us with the potential of the device. But how they are designed and the results (how they sound) are in the end, what really matters.

Thanks for the post.

 

I almost totally agree.

The one thing I am not so sure about is even resampling vs uneven resampling. I am still doing some tests but one thing I just found yesterday. You can actually brute force a simple SRC like soundforge into performing almost as good as the players - SoX, R8Brain, iZotopes. You just have to do it in steps and use some math though. We used to do something similar with photos and photoshop because we didn't trust the algorithms for certain operations.

Anyway the problem I see with uneven resampling is a possible trade off. In some scenarios I could see people trading off low level measurable distortions that can not be heard for hard to measure phase distortions that are audible (ringing).

 

I think it's important to remember that when most people make a categorical statement they are talking about the rule rather than the exception.

When someone says "analog EQ is better than digital EQ" I think it's a more reasonable conclusion to interpret the statement as shorthand for "in my experience if I run a signal through an analog EQ with which I am unfamiliar the chances are greater that I would be happy than if I run the same signal through a digital EQ with which I am unfamiliar". I don't think the person really means to say "it is impossible for a digital EQ to be as good as an analog EQ".

But lets talk about verb for a second rather than EQ - because it goes with the discussion. Almost all the reverb you get in studios these days is digital. It has improved a lot. I can hear a really great recording with digital reverb on it. It has NEVER, in my experience, not been a LOT more work than doing the mix than it would be with a plate or a chamber. But that's not meant as a categorical statement, just a reflection of my experience...

 

Thanks Barry! The voice of reason and experience once again.

We keep trying to say "It's the results that matter", but often that gets flushed down the drain here. What it sounds like to each of us is the only final grade that means anything to me. If I can't hear a difference or improvement, then for me only it just isn't there. Implementation gives us these differences and allows for the discussion of our Opinion of what sounds better.

Hopefully this will help inform future conversations about these critical sound issues. Very informative as always. Thanks for the continued insight!!!

Thank you for continuing to post here! I find your shared thoughts to be invaluable.

 

Hi Key,

My criteria for assessing something like a sample rate conversion algorithm is direct comparison against the unconverted original.

I would be curious as to how you can "brute force" something like SoundForge's SRC "into performing almost as good as" iZotope's. Perhaps the definition of "almost" comes into play. I wonder how you'd get rid of the aliasing and spurious harmonics present in the former and absent in the latter.

On my system, every SRC algorithm sounds very different from every other one. They aren't close.

There is no "trick" to well done non-integer conversion. It is simply more "complex" math, that takes the file to a common denominator.

Finally, to test integer vs. non-integer, take a great source and record it say, different high sample rates, say 96k and 88.2k. Start with these two source files and use each of a bunch of SRC algorithms to create 44.1k versions. Compare all the 44.1k files with the original, unconverted files.

So far, I've done this with two dozen algorithms. The more transparent ones don't "care" whether the conversion is integer or non-integer. It is only an issue for the ones that "can't do the math". And those tend to be artifact ridden anyway, so I would not use them in the first place.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Well yeah I think reverb is just hard as hell to simulate. To my ears the only ones that approach realism beyond the real thing miced or maybe a plate are impulse response based reverbs. Pretty much everything else sounds like an artifice or an abstract representation of reverb.

Also it depends on how it's mixed. If you mix a digital eq to the point where it's felt more than it is heard then I generally don't get the "fake reverb" red flag popping off in my head. But of course you can really turn up the volume on the better reverbs and it wont sound fake so maybe this is just a workaround.

 

For reasonable people, yes. But I’m just not sure that most people are reasonable!

 

the hard part for me is the beauty of blending different sounds in the plate. that's where most digital verbs fall apart in my experience. (note all the qualifiers!)

 

Hi Steve,

I made this thread and put up the initial post precisely because I have seen many sweeping generalizations, not reports limited to personal experience and in many cases with no experience at all.

Folks criticize a release because they heard somewhere it used "digital EQ" or "digital limiting" etc.. Others think a recording will necessarily be wonderful because the recording chain used tubes.

As to hearing digital reverb, I'm sure audibility as such is true with a lot of digital reverb (or how it was used), just as many (but certainly not all) digital EQs will sound "digital". I'm equally sure that not all digital reverb is identifiable so easily.

Just my perspective.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Well the key part of the "almost" is possible user error on my part. I am not the best at math haha and I was just amazed that I got it to work at all so I haven't figured out if I am doing it the best way or not.

Also because I am misusing the SRC to an extent I might be fooling the program into thinking the nyquist frequency is at a much higher location and I might be getting some aliasing as a result. I can fix that easily by placing a vst antialias filter in the chain right before the last downsample.

Here are the screen shots. Big thanks to Neil for the theory used and a big thanks to Moon Man for going back and forth with me figuring out the specific math needed - and basically doing the math for half of it.

192kHz into 44.1 kHz (24-bit)
Soundforge normal resample (High level 4 with antialias filter)


192kHz into 44.1 kHz (24-bit)
Soundforge brute force stepped resample (High level 4 antialias filter)


With the second one I did 192k x14, x10.5, /5, /8, /16 with a batch conversion I came up with.

 

Love that analogy! I think it applies to audio beautifully, as many components (or recordings) are much more than the "sum of the parts". High quality parts implemented poorly can be much worse than a clever implementation using modest parts.

 

Barry, do you have any digital EQ software recommendations on the PC side?

 

Hi Key,

As far as I can see from these graphs, neither of those sweeps is as clean as some of the better SRC algorithms can do. And the second ("forced") graph doesn't look as clean as the first one. So far, it isn't "almost" like iZotope from what I can see - unless we define "almost" very differently.

Do you have iZotope's SRC?
Take a good high res original and convert it to 44.1k using SoundForge at its best setting and do the same using iZotope's SRC. Compare both to the high res original. Do they sound the same (or even close) to your ears?
To mine, they are worlds apart.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Hi Shawn,

I'm sorry I don't.
I've heard some nice ones, such as the EQ available in WaveLab (but my favorites of all I've heard so far have all been Mac based).

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com