The No.1 Website for Pro Audio
Lets do it: The Ultimate Plugin Analysis Thread
Old 3rd March 2019
  #3451
Deleted 691ca21
Guest
Quote:
Originally Posted by Arksun View Post
Ahh, when he said truncating I thought he meant to 24-bit fixed, because that would make sense as an argument.

But you say truncating to 32-bit float?

Wow, just wow, who knew below -180dbfs was so important

Perhaps when humans have evolved to clearly hear a dynamic range from a whisper to a thermonuclear explosion and such speaker technology exists to churn out 180db levels with almost 0 distortion levels, we can start to get concerned about it then
The errors can accumulate and possibly become audible, there are a few threads discussing this on here recently.
Old 3rd March 2019
  #3452
Lives for gear
 
Arksun's Avatar
Quote:
Originally Posted by Deleted 691ca21 View Post
The errors can accumulate and possibly become audible, there are a few threads discussing this on here recently.
Even after accumulation of 500 tracks, error levels in mathematics well below the audible boundaries of the physical universe aren't really that big a concern are they?

Sure, in say a reverb plugin which is highly iterative processing, 64-bit internally makes sense. But a 32-bit float eq plugin in a 32-bit float mixer having a 32-bit floating noise floor is a total non-issue.

Of all the aspects elements and algorithms that can affect how a piece of music sounds working digitally, this is so low down on the list as to be laughable.
Old 3rd March 2019
  #3453
Gear Addict
 

Quote:
Originally Posted by Arksun View Post
Even after accumulation of 500 tracks, error levels in mathematics well below the audible boundaries of the physical universe aren't really that big a concern are they?

Sure, in say a reverb plugin which is highly iterative processing, 64-bit internally makes sense. But a 32-bit float eq plugin in a 32-bit float mixer having a 32-bit floating noise floor is a total non-issue.

Of all the aspects elements and algorithms that can affect how a piece of music sounds working digitally, this is so low down on the list as to be laughable.
How do you came with the number 500 tracks?

What is the scientific basis on this? Facts? Links?

I suppose Cubase Pro 10 64 bit processing was made for a reason.
Old 3rd March 2019
  #3454
64bit I/O support is "made" by replacing the word "float" into "double". A matter of 3 sec. 64bit was not made by Steinberg, it's essentially free for modern CPUs.

Qualitative benefit for the end user is zero. But he feels better, and that doesn't hurt anybody.
Old 3rd March 2019
  #3455
Gear Addict
 

Quote:
Originally Posted by FabienTDR View Post
64bit I/O support is "made" by replacing the word "float" into "double". A matter of 3 sec. 64bit was not made by Steinberg, it's essentially free for modern CPUs.

Qualitative benefit for the end user is zero. But he feels better, and that doesn't hurt anybody.
64 bit for mixing - no floating point noise from errors. 64 bit for I/O - no loss of info when transferring between DAWs. This is basic level 101 in digital world.

It is common sense we should use solution with minimum errors and loss of info if possible. Bit errors add up in the digital world and should be minimized when working with DAWs.
Old 3rd March 2019
  #3456
So why don't we try 2048bit numbers? Any cheap browser uses them.

It's irrelevant, only reason we see 64bit audio busses here and there these days is because the move is almost free of cost. But it is incredibly difficult to demonstrate any necessity for such hilarious precisions.

Your own presence affects the trajectory of Jupiter, by some amount. Only question is, where is the benefit of taking this info into account in your daily life? It's zero, almost. A very similar zero to the one I mentioned above.

One simple DA -> AD insert will introduce a several thousand times greater error, every additional processor in the chain, too. See this in relation for a moment! Audio plugin I/O is by miles the smallest, the most irrelevant problem you'll find in digital audio today.



Knowing when to stop and why, is a quality (not just in engineering)

A psychologist would even argue that, not doing so, at some point, will start to hurt.

Last edited by FabienTDR; 3rd March 2019 at 10:51 PM..
Old 4th March 2019
  #3457
Lives for gear
Quote:
Originally Posted by 36936 View Post
64 bit for mixing - no floating point noise from errors. 64 bit for I/O - no loss of info when transferring between DAWs. This is basic level 101 in digital world.

It is common sense we should use solution with minimum errors and loss of info if possible. Bit errors add up in the digital world and should be minimized when working with DAWs.
But can you prove it with a null test?
Old 4th March 2019
  #3458
Lives for gear
Quote:
Originally Posted by miscend View Post
But can you prove it with a null test?
"it" here meaning what specifically? That bit error buildup is real in non-64bit environments?

Even if you do get a non-null what would that prove?
Old 4th March 2019
  #3459
Lives for gear
Quote:
Originally Posted by Mikael B View Post
"it" here meaning what specifically? That bit error buildup is real in non-64bit environments?

Even if you do get a non-null what would that prove?
It would prove existence of errors possibly in form of noise.
Old 10th May 2019
  #3460
Gear Addict
 

Anybody tested the Mia Compressor One yet? I did some tests and found some surprising results that I think some people would find very interesting. I can post some pics later today but it's a compressor with 5 analog models and I'll just say that they in no way show up in testing how they should. I'm no expert and maybe there is something I'm missing the but I was very surprised with the results.
Old 13th May 2019
  #3461
Lives for gear
Quote:
Originally Posted by FabienTDR View Post
One simple DA -> AD insert will introduce a several thousand times greater error, every additional processor in the chain, too. See this in relation for a moment! Audio plugin I/O is by miles the smallest, the most irrelevant problem you'll find in digital audio today.
Fabien, I really appreciate your expertise here. I have a question regarding your statement above: Do you think that these errors should be avoided by not using external gear as inserts for processing? I was considering buying some rack mount hardware to add saturation to tracks I have already recorded, but these errors introduced by going through a round of DA - AD have me concerned.
Old 13th May 2019
  #3462
Deleted 691ca21
Guest
Depends if those errors are audible (to you). I still prefer a DAC-Analogue Chain-ADC loop, because it simply sounds better (to me) than plugins, especially for vibe/saturation.
Old 13th May 2019
  #3463
Gear Guru
Isn't the reason for 64 bit to optimize plug in performance? I do agree that all the bit depth sample rate stuff seems to be a non issue audibly, (unless working on half speed efx)..... One guy was telling me if you want to CYA and have huge files by all means have fun!.....You most likely won't hear a difference....

His bigger point was for my bedroom projects it'd be cologne on a pig.......! He also said if a client wants it he won't argue......
Old 13th May 2019
  #3464
Lives for gear
 
b0se's Avatar
Quote:
Originally Posted by Notheorem729 View Post
Anybody tested the Mia Compressor One yet? I did some tests and found some surprising results that I think some people would find very interesting. I can post some pics later today but it's a compressor with 5 analog models and I'll just say that they in no way show up in testing how they should. I'm no expert and maybe there is something I'm missing the but I was very surprised with the results.
Interest piqued, care to elaborate?
Old 13th May 2019
  #3465
Gear Addict
 

Quote:
Originally Posted by b0se View Post
Interest piqued, care to elaborate?
Yeah unfortunately it's far too much to type but I have a video I recorded that I'm posting later.

To summarize... its supposed to emulate analog gear and has the modes class a, fet, opto, tube 1, and tube 2. However, the dynamic functionality and harmonics don't change between the modes but the frequency spectrum does. It applies some crazy EQ curves most of which you can hear and most of which have crazy sharp peaks that introduce resonance. It seems like they applied matching EQ more so than emulating dynamic behavior. Strangely enough nobody seems to have noticed.
Old 13th May 2019
  #3466
Lives for gear
 
b0se's Avatar
Quote:
Originally Posted by Notheorem729 View Post
Yeah unfortunately it's far too much to type but I have a video I recorded that I'm posting later.

To summarize... its supposed to emulate analog gear and has the modes class a, fet, opto, tube 1, and tube 2. However, the dynamic functionality and harmonics don't change between the modes but the frequency spectrum does. It applies some crazy EQ curves most of which you can hear and most of which have crazy sharp peaks that introduce resonance. It seems like they applied matching EQ more so than emulating dynamic behavior. Strangely enough nobody seems to have noticed.
Ah I see. I've only tested it with the modes disabled for compression only. Just an EQ change doesn't sound good.
Old 13th May 2019
  #3467
Gear Addict
 

Quote:
Originally Posted by b0se View Post
Ah I see. I've only tested it with the modes disabled for compression only. Just an EQ change doesn't sound good.
I wouldn't care if it wasnt for the fact that the curves are there even with no compression. Plus the curves are very bad. very sharp cuts and boosts. Try tube 2 on bass.
Old 13th May 2019
  #3468
Lives for gear
 
b0se's Avatar
Quote:
Originally Posted by Notheorem729 View Post
I wouldn't care if it wasnt for the fact that the curves are there even with no compression. Plus the curves are very bad. very sharp cuts and boosts. Try tube 2 on bass.
Just had a play with the modes and wasn't impressed with the sound so opened up Plugin Doctor to check - pretty extreme curves. Passing on this one anyway, doesn't offer anything novel.
Old 13th May 2019
  #3469
Gear Addict
 

Quote:
Originally Posted by b0se View Post
Just had a play with the modes and wasn't impressed with the sound so opened up Plugin Doctor to check - pretty extreme curves. Passing on this one anyway, doesn't offer anything novel.
Yeah same here. I'ts a decent compressor until you turn on the simulations then it falls apart. I started taking pics to post here but with 6 modes it ended up being so much I decided to just make a video showing them with audio in the background. It's very bizarre because of how many people haven't noticed because some of the resonances it introduces are so bad and obvious that I can't believe nobody noticed them.
Old 14th May 2019
  #3470
Lives for gear
Quote:
Originally Posted by Notheorem729 View Post
I wouldn't care if it wasnt for the fact that the curves are there even with no compression. Plus the curves are very bad. very sharp cuts and boosts. Try tube 2 on bass.
Maybe these curves are indeed "the distinctive sound of MIA Laboratories"?
Old 14th May 2019
  #3471
Gear Addict
 

Quote:
Originally Posted by Mikael B View Post
Maybe these curves are indeed "the distinctive sound of MIA Laboratories"?
I sure hope not. We're talking about some serious curves. Boosts and cuts that exceed 6db in some cases and are sharp enough to almost start self oscillating on many sources. Some of the modes shift the entire spectrum.
Old 1st October 2019
  #3472
Lives for gear
 
Pali's Avatar
 

Sorry if this has been mentioned before but how can one analyse an internal side chain. Like the response of the API 2500 Thrust circuit.
Old 15th October 2019
  #3473
Gear Nut
 

I'm curious, is there anything technically wrong with using waveshapers moderately to create saturation ITB?
Old 15th October 2019
  #3474
Here for the gear
 

Quote:
Originally Posted by Arksun View Post
Even after accumulation of 500 tracks, error levels in mathematics well below the audible boundaries of the physical universe aren't really that big a concern are they?

Sure, in say a reverb plugin which is highly iterative processing, 64-bit internally makes sense. But a 32-bit float eq plugin in a 32-bit float mixer having a 32-bit floating noise floor is a total non-issue.

Of all the aspects elements and algorithms that can affect how a piece of music sounds working digitally, this is so low down on the list as to be laughable.
First of all, a 32 bit floating point do *not* have 32 bit noise floor, but 24 bit.
The caracteristic of the floating point representation is that noise floor and dynamics are separated; dynamic is handled by the mantissa, the 8 missing bit.

Then, adding 512 signals (512 is simpler than 500 in the following) need 9 bit of headroom; you see, you are already handling your signal with 15 bits, not 32.

Then, you do things with signals; in general linear operations, like filters, equalisation, etc. You apply each of this operation also to the errors (noise floor) you have in the signal; some computation have the inelegant property of amplify the error (**).

While it is possible to reduce errors with carefully designed algorithms and code, i would not say that 64 bit is used only for convenience, it surely help the design of DSP algorithms including those used in a mixer and avoid problems.

Maurizio

(**) the subject of errors in floating point computation is complex, it is a university courses subject; it is not al all intuitive, and it is connected to how the algorithm is written, which operations are done and in which orders ; there limit cases, for exemple using a standard linear algorithm to inverse the wrong matrix, where you can easily obtains errors around 10% using floats or even doubles.
Old 16th October 2019
  #3475
Cytomic
 
andy-cytomic's Avatar
 

Quote:
Originally Posted by maurizio.dececco View Post
...

Maurizio

(**) the subject of errors in floating point computation is complex, it is a university courses subject; it is not al all intuitive, and it is connected to how the algorithm is written, which operations are done and in which orders ; there limit cases, for exemple using a standard linear algorithm to inverse the wrong matrix, where you can easily obtains errors around 10% using floats or even doubles.
Very good point there, I've never been a fan of software that states "double precision used throughout" since it sounds to me like the people writing it are crossing their fingers that using it will avoid issues. To me that's just a waste of compute power and doesn't guarantee that numerically poor algorithms will do any better anyway. I make judicious use of double precision, but only where it is needed and otherwise use single precision since it's faster and sounds just as good!
Old 16th October 2019
  #3476
Cytomic
 
andy-cytomic's Avatar
 

Quote:
Originally Posted by TimPani View Post
I'm curious, is there anything technically wrong with using waveshapers moderately to create saturation ITB?
If you are trying to get some sort of subtle analog saturation sound then you need to model the analog circuit you like the sound of. Most circuits cannot be modelled with trivial waveshapers. Circuits balance the current travelling in and out of every node, and typically have non-linear voltage to current relationships, not non-linear voltage to voltage ones that you get from a trivial digital waveshaper. But, ignoring analog modelling, if you want to add some subtle distortion with a waveshaper then you need to either oversample, or use other methods to reduce aliasing generated.
Old 16th October 2019
  #3477
Here for the gear
 

Quote:
Originally Posted by andy-cytomic View Post
To me that's just a waste of compute power and doesn't guarantee that numerically poor algorithms will do any better anyway. I make judicious use of double precision, but only where it is needed and otherwise use single precision since it's faster and sounds just as good!
True. I just have a question (a real one, not rethoric : is still true that single precision is faster than double ?

I was true at my 'good old times', old architecture where a double computation was decomposed in multiple single precision computation, but is it still true with recent Intel/ARM processors ? If the floating point hardware is double precision, the speed difference may not exists any more, or being very reduced. Any actual data ?


Maurizio
Old 16th October 2019
  #3478
Cytomic
 
andy-cytomic's Avatar
 

Quote:
Originally Posted by maurizio.dececco View Post
True. I just have a question (a real one, not rethoric : is still true that single precision is faster than double ?

I was true at my 'good old times', old architecture where a double computation was decomposed in multiple single precision computation, but is it still true with recent Intel/ARM processors ? If the floating point hardware is double precision, the speed difference may not exists any more, or being very reduced. Any actual data ?


Maurizio
It depends on the platform. On Intel modern scalar (one number at a time) single vs double is the same speed using the same size registers. On Intel at the lowest level of SSE you have 4x single SIMD vs 2x double SIMD. So you get to process twice as many numbers in single vs double precision.

I'm not sure on ARM NEON what the story is, but I'll be exploring that space in the next few years. I'm pretty sure on the most recent 64-bit chips is pretty similar to SSE Intel, but it may be that double is a little slower since its early days for ARM doing double precision SIMD. If any devs know please pipe up!
Old 18th October 2019
  #3479
Airwindows
 
chrisj's Avatar
Quote:
Originally Posted by andy-cytomic View Post
But, ignoring analog modelling, if you want to add some subtle distortion with a waveshaper then you need to either oversample, or use other methods to reduce aliasing generated.
It's also hugely dependent on the style of waveshaping and your style of music content generating. Very subject to what you're doing with it.

I'm pretty much the poster child for getting a sound through minimal processing of waveshaping with the gentlest possible transfer functions. This has the following effects:

-any frequency subject to this waveshaper is going to produce a third harmonic well before it produces anything else. That's dependent on how 'soft' the curve is: if there are any discontinuities, 'analog-accurate' or not, you immediately have more aliasing.

-any frequency over half of Nyquist will immediately produce aliasing. At CD quality, that's 11.025K (already quite high). At 96K, it's 24K. Pure frequencies below that, hitting the waveshaping gently, will NOT be producing significant aliasing until they're distorting so hard they're producing a fifth harmonic (or indeed fourth) while also being above the problem frequency… roughly 5K, or at 96K it'd be 12K while distorting so hard it produces that fifth harmonic out of a gentle waveshaping.

Some folks are constructing sounds ITB with an eye to brightness, and lots of their audio is over 11K, and if they use naive waveshaping they'll immediately get obvious aliasing.

Others are recording analog, or seeking mellower tones, and of course AMPLITUDE is everything here. If your 74 hz bass note is driving the waveshaping hard, but its higher partials are all very weak, well the aliasing will also be weak or even nonexistent. You can also prefilter before doing the naive waveshaping, simply throw out the trouble frequencies if you know you're going to be avoiding them later in the chain. Odds are you can't be doing that with your cymbals or lead vocals with bright esses…

It all depends. Some people absolutely cannot use naive waveshaping. For others, it's really not such a big deal (and this is entirely outside of aliasing as a gritty sonic choice, which is also a possibility… but not one you want to stumble into by accident).
Old 5th November 2019
  #3480
Gear Head
 

Say whaaat?

I just discovered this thread and still can't believe I didn't know this existed. Still my now "dirty" Sonnox is my favorite clean comp, I may have to dive deeper into this issue or forget about it as quickly as I can
Of course I started to play around, checked some plugins and also had a look at my little hardware setup to confirm how harmonics look in the real world. No surprises there, 76-KT - even and odd harmonics, SSL Six - way less even and odd harmonics, GBus Comp - more even and odd harmonics. Then I tested the channel comp... Shouldn't that be impossible? How can that be? Are they secretly using a DSP? Do I have to make sure that my little analog desk doesn't suffer from digital aliasing artifacts these days?

Two screenshots - one with the sine running through the channel clean and one where I engaged the comp. Nothing else changed, I am not clipping the converter or anything like that. What's going on?
(Still like the sound though )
Attached Thumbnails
Lets do it: The Ultimate Plugin Analysis Thread-ssl_six_nocomp.jpg   Lets do it: The Ultimate Plugin Analysis Thread-ssl_six_channelcomp.jpg  
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump