What are some central ways in which DSP plug-ins are made to sound different, even if the basic algorithm is common?

[soundmodel]

What are some central ways in which DSP plug-ins are made to sound different, even if the basic algorithm is common?

Like, lets assume that most EQ plug-ins would actually use RBJ filters. So if they sound different, then how?

Or lets consider choruses. The basic idea is the same, so how do they sound different, if they do?

Are the differences only superficial? Like different parameter ranges.

[Music Engineer]

In a chorus, you have to make a couple of design choices: number of voices, delay times, interpolation method, waveforms/frequencies/phases of the modulators - and maybe more. Each of these choices will affect the sound. RBJ filters should indeed sound equal - unless the implementor did something wrong. There might be subtle differences in roundoff behavior depending on which of the 4 different direct forms is used (if any) - but I doubt that they are audible (Edit: I never used them in single precision though, so I might be wrong - oh - ah - yeah - now I notice: you can use single or double precision and maybe that matters in certain contexts - and some people may even use more exotic number formats like fixed point). However - they may behave/feel different depending on the chosen parametrization for specifying the bandwidth in case of bandpass/notch/peak (via Q or in octaves, say). Oh - and I'm assuming fixed settings. As soon as modulation comes into play, the implementation structure does matter as well.

[BertKoor]

"It depends" because each design / implementation decision is a trade-off

[soundmodel]

"It depends" because each design / implementation decision is a trade-off

In what sense?

How are RBJ filter (or SVF) -based plug-ins tradeoffs? Well, it would be e.g. if one uses it to implement a Pultec-clone instead of some 20-band monster? However, these would be sonically identical, right?

[ghettosynth]

Wooden end cheeks.

[BackInCheck]

Well, you can make your EQ plugin sound "unique", despite using standard filters, by choosing specific parameters and parameter ranges that are being exposed to the users, as opposed to presenting all potential parameters and full parameter ranges. That's why many swear on their favourite channel / analog - inspired EQ, that often offers less than a fully parametric EQ, but more of what the user actually needs / wants.

For example: a Pultec-inspired EQ plugin can use standard filters under the hood, but it will behave different from a generic parametric EQ, since you only have a certain number of bands and frequencies to choose from and bandwidth (Q) can only be changed on one band. The "boost" and "attenuate" settings will work on slightly different frequency points, even if the band is set to the same value.

This can all be achieved with standard filters, that are deliberately implemented to recreate the behaviour of the Pultec. So, user interface + choice of parameters & parameter ranges makes an EQ plugin unique.

A different approach is to implement some saturation / waveshaping stages (at the signal input, output or even in the EQ bands itself) to spice up the standard filters.

I'm not a programmer myself, but from my observation, this seems to be the main thing that distinguishes "analog style" EQ plugins, I think.

[gambero]

Joke answer: graphics
Real answer: basically the sum of all previous answers.
Different developers take different pathways to the same product, and since there's no real perfect solution for making the best sounding eq, it comes down to (as BertKoor said) choosing the right tradeoff of pros and cons, and personal opinion. Parameter ranges do affect the overall "sound" of a product quite a lot.
Keep in mind that many standard algorithms are very often used as a starting point for modifications, expecially for adding nonlinear behaviour.

[soundmodel]

For example: a Pultec-inspired EQ plugin can use standard filters under the hood, but it will behave different from a generic parametric EQ, since you only have a certain number of bands and frequencies to choose from and bandwidth (Q) can only be changed on one band. The "boost" and "attenuate" settings will work on slightly different frequency points, even if the band is set to the same value.

Oh yes, this is a good point indeed. I forgot that when you combine different basic filters at different gains, you can also produce wildly different curves for the same UI-based gains or cutoffs.

E.g. mix a peak and a high-shelf, but in proportions so that the peak resonates the cutoff point and possibly reacts to gain non-linearly.

[BertKoor]

How are RBJ filter (or SVF) -based plug-ins tradeoffs?

This thread plus some others you started make me believe you see DSP as a SMOP.

Different implementations of the same textbook algorithm should indeed sound identical. And they will be identical within the constraints in which they behave linear. But linear is boring. Linear is not musical.

Introduce non-linearities into the equation, and all bets are off.

In school I had some basic electrical engineering. Most exam questions started with "you may assume the voltage source is ideal". Well, in practice no voltage source is ever ideal.

Deviation from "ideal" or "normal" is the definition of "character".

Do you know what a transistor does? In theory it is simple. But in practice: no transistor is ideal. There are tens of thousands types of transistors, all behaving slightly diffrrently. How does an engineer pick the right one for the job? It's not my job (I became a sofware engineer, not electrical engineer) but I know it's always a compromise.

All engineers strive for perfection, but that is not a realistical goal.

So... 40 years ago I was in the market for a flanger. Spent an hour in a music shop and went home with the one I liked most. A friend also wanted a flanger and did the same. He bought the one I liked the least! While their operating principles and controls were exactly the same, they had different characters. How? I guess the engineers cut different corners.

You always need to cut corners. Decisions, decisions...

So I bought an analog bucket brigade delay. Max delay time: 350 ms. Some years later I bought a "better" one: max delay time: 400 ms. It's delays also sounded more faithful to the original signal. Years later I got a digital delay. But guess which delay pedal I later sold off and which one I held on to, for it's character...

I have taken a look inside effect pedals more than once. They contain dozens and dozens, perhaps even hundreds of transistors and capacitors. How does the engineer pick the right ones you think? None are ideal, it's always a compromise.

I think some characteristics often are not the result of conscious design decisions, more the result of happy accidents. And we only come to appreciate them when they are missing. It's really weird. I don't think there is a recipe for good or bad, time will tell.

So this complex world of analog gear we try to emulate with DSP. And that comes with its own problems, challenges and headaches. Resampling a signal for instance. In principle it is simple... until you try it.

Go try it!

Analog schematics often have feedback paths. In software that is a real problem. Go try it...

[soundmodel]

^ Yes so does your answer support the idea that the grit of this issue is not in standard algorithms, but how they're applied?

When it comes to EQs, then a paper demonstrated that the SVF is not only numerically nice, but also good-sounding.

EDIT: this paper: http://dafx14.fau.de/papers/dafx14_aaro ... s_for_.pdf

[BertKoor]

Do you believe everything written down?
I'll just repeat: go try it!

[G-Spot]

There is plenty of neglected design space for complicated and/or computationally intensive processing.
Most software sticks to simple and efficient basic boring algorithms, even if there is no need to, because the theory is

• usually intended to optimize for the cheapest hardware that can do an adequate job (e.g. fixed point numbers, low sampling rates, short delay lines)
• overly focused on easy cases, first and foremost linear systems and memoryless nonlinearities

Leaving this comfort zone requires ability to do original research, and creative ideas to direct it; not by chance, traditional heavyweight plugin types like convolution reverbs and massive additive synths are straightforward and generic.

[soundmodel]

There is plenty of neglected design space for complicated and/or computationally intensive processing.

Yes, so like U-he Diva?

[whyterabbyt]

^ Yes so does your answer support the idea that the grit of this issue is not in standard algorithms, but how they're applied?

When it comes to EQs, then a paper demonstrated that the SVF is not only numerically nice, but also the most good-sounding.

which paper?

[G-Spot]

There is plenty of neglected design space for complicated and/or computationally intensive processing.

Yes, so like U-he Diva?

U-he Diva seems very well designed, but it is complex because it consolidates and accumulates many traditional features of vintage hardware and allows to use them together, not because of obvious novel designs. An assortment of filters and oscillators that sound good is certainly a more creative design than pedantically emulating specific hardware, but it is still a modest step forward.