Is Gullfoss any good?

[Ploki]

Gullfoss is a FFT based EQ, it's technically "a lot of bands" (probably 256-2048, more you get time-domain smearing) but not really because it's a different process than having split bands.
It's either linear phase, or the phase changes with the amount of processing, like with dynamic EQs.

That's not accurate. Gullfoss does not use FFTs for modifying the signal. It employs a proprietary serial filter bank that has unique properties when it comes to avoiding artefacts, preserving temporal coherence, allowing for ultra-fast reactions and highly accurate control over amplitude and phase. In fact, this part of Gullfoss is partially covered by a Soundtheory patent.

Your assertions about FFT processing introducing time-domain smearing also only apply to the most naïve use of spectral filtering, i.e. modifying the short-time Fourier spectrum and directly resynthesizing it with an inverse FFT and producing temporal aliasing this way. A properly implemented FFT based convolution algorithm produces the exact same result as a time-domain convolution, only faster.

TEOTE is a "fixed filter bank" - it's made of actual crossovers (3-64), not linear phase, so it has a bit of phase distortion - but its CONSTANT and its the same no matter the amount of processing. if you like how it sounds when it does NOTHING, it won't f**k phase when it does something.

That's not correct. Splitting a signal into parallel bands that are then individually amplified and summed together does not in general result in a static phase response, even if the filter bands do have a static response. Consider any frequency between two bands where both band responses contribute with roughly the same amplitude but different phases. The total response at this frequency is, therefore, the scaled sum of two complex numbers with different phases, as in a_low(t) * h_low(f) + a_high(t) * h_high(f) where a_low, a_high are the band amplifications at time t and h_low and h_high are the band responses at the frequency f. You'll notice that the argument of this sum arg(a_low(t) * h_low(f) + a_high(t) * h_high(f)) depends on a_low and a_high in general, and hence on the time t.

As a consequence, your parallel filter bank not only introduces phase error in between the bands upon reconstruction, it also has a time-varying phase response wherever the bands overlap and such a time dependence introduces uncontrolled and time-depending temporal signal dispersion at the cost of temporal coherence.

Gullfoss on the other hand will change phase response as its processing, wreaking havoc to phase and transients. Unless you use linear phase, which has pre-ringing.

It's correct that Gullfoss uses a non-constant phase response, but your conclusion is incorrect. Gullfoss' phase response is well controlled to the effect of improving temporal coherence on average. It also avoids pre- and post- ringing by understanding how exactly a signal can be modified while not changing the grouping of auditory elements, which means you don't get sounds that were not there before, preventing any ringing or other artefacts from being introduced.

If you had an issue with transients when using Gullfoss, it may be that you were observing a bug that got fixed a while ago that had some impact on signal integrity. We've also introduced Gullfoss Live, which does not process transients at all, if that is what you are looking for. But in general, if used correctly, Gullfoss does not damage signal transients.

Thanks,

Andreas

I stand corrected! Thanks taking your time to clarify.
I’ll give it another spin then, looks like I sold it before the bug with transients was fixed, i was assuming it was a processing side effect.

How come the phase response changes with magnitude of processing if it employs a fixed filter bank?
Also is the patent public (curiosity nothing more).

Is it a secret how many bands gullfoss uses internally?

Edit:
If someone missed it, there’s a bunch of good answers to other people from atell/Andreas on the previous page

[plexuss]

Thanks atell for correcting the fallacies people posted. I wish people would stop posting things they havent validated to be true. When I read stuff like that I unfortunately have to ignore it because the poster doesn't provide any validation of their post. Its very valuable when some who actually knows the facts shares them. Thanks again!

So Gullfoss is based on a real-life dynamic waterfall software model using AI and batteries from Tesla!

[atell]

How come the phase response changes with magnitude of processing if it employs a fixed filter bank?
Also is the patent public (curiosity nothing more).

If you look at each band individually, you get a static response. But because the individual bands don't agree on a phase for certain frequencies, if you start to amplify and mix the bands, you get phase interactions between the static contributions that depend on the band amplifications. It's not that to visualize if you work it out on a piece of paper.

Here's the Soundtheory patent:
https://patentimages.storage.googleapis ... 158341.pdf

Don't get confused by the figure with the triangular bands. Those are not the filters. It's a simple example for a set of basis functions that can be used to describe the target frequency response that is then approximated by the actual filters. The patent describes the general approach we use for Gullfoss, but leaves out a number of important components, so you won't be able to replicate our results with only this information.

Is it a secret how many bands gullfoss uses internally?

It's not really a secret, but it's also not technically a band count. We use up to 1152 discrete filters per channel. Those are not bands, though, i.e. they don't split the signal into that number of (approximately) bandlimited sub-signals. But their combined effect is to give the frequency magnitude response displayed by Gullfoss and without any cross-over effects.

Thanks,

Andreas

[MogwaiBoy]

Gullfoss does not use a fixed target curve or anything like that. The processing decisions it makes are entirely context-dependent and are trying to maximise the audible information in the signal. That means that Gullfoss tries to prevent individual signal components from masking each other. The rebalancing effect emerges from this approach and it is not guided towards some kind of optimal global frequency distribution. That's why the Brighten control is very important. It gives Gullfoss a hint regarding the desired power distribution in the signal. Local avoidance of masking does not fully determine the signal distribution. If you have two separate tones, then there is a wide range of amplitudes for the two tones for which they are not masking each other. You can make the higher frequency tone louder than the lower frequency one, or the other way around, and still hear both. The Brighten parameter gives you control over precisely this degree of freedom without imposing a general slope for the frequency distribution. So the result depends on what is already present in the signal, as it should be. Gullfoss is not a dynamic matching EQ, unlike most of the competing plugins.

It's not FFT, it's not fundamentally a dynamic EQ, it's not minimum phase, it's not linear phase - what is this sorcery

[plexuss]

Gullfoss does not use a fixed target curve or anything like that. The processing decisions it makes are entirely context-dependent and are trying to maximise the audible information in the signal. That means that Gullfoss tries to prevent individual signal components from masking each other. The rebalancing effect emerges from this approach and it is not guided towards some kind of optimal global frequency distribution. That's why the Brighten control is very important. It gives Gullfoss a hint regarding the desired power distribution in the signal. Local avoidance of masking does not fully determine the signal distribution. If you have two separate tones, then there is a wide range of amplitudes for the two tones for which they are not masking each other. You can make the higher frequency tone louder than the lower frequency one, or the other way around, and still hear both. The Brighten parameter gives you control over precisely this degree of freedom without imposing a general slope for the frequency distribution. So the result depends on what is already present in the signal, as it should be. Gullfoss is not a dynamic matching EQ, unlike most of the competing plugins.

It's not FFT, it's not fundamentally a dynamic EQ, it's not minimum phase, it's not linear phase - what is this sorcery

It's sorcery, exactly. And exactly why I use Gullfoss. I think end users think that plugins are built from pre-existing libraries of functionality and in essence "all plugins are the same". Sure devs use pre-existing libraries and techniques to pull off their intent. However, there are devs that go beyond and work out their own unique approach to functionality. This, to me, is the interesting part of the competition between audio sofware products. Some devs will use pre-existing code only and they will end up with a more generically sounding product. Other devs put effort into innovation and some of them succeed in making a truly great sounding unique product. This is where I think Gullfoss is with their product. At least, that's what it sounds like to me. atell's details here further validate my end-user experience and assumptions (based on listening only) so thanks for that!

[Ploki]

I wasn't talking about pre-existing libraries, but there's a limited amount of methods you can use to process a signal, simple as that.

[plexuss]

I wasn't talking about pre-existing libraries, but there's a limited amount of methods you can use to process a signal, simple as that.

I think that assumtion is fundamentally incorrect. It's code - there are almost an unlimited amount of methods that can be used to process a signal.

[Azbest]

I'm demoing this right now. It sure does like bright mixes. I don't know if I'm happy about it.

edit
After some brief testing I think this plugin could benefit from an option to limit the range it works with. I like what it does to lows and mids, but I like my top end a bit subdued, and Gullfoss makes it too harsh for my taste. I have to apply an EQ after it to keep the piercing sizzle at bay. Maybe I just suck at using it yet, will do some more testing.

[plexuss]

I'm demoing this right now. It sure does like bright mixes. I don't know if I'm happy about it.

edit
After some brief testing I think this plugin could benefit from an option to limit the range it works with. I like what it does to lows and mids, but I like my top end a bit subdued, and Gullfoss makes it too harsh for my taste. I have to apply an EQ after it to keep the piercing sizzle at bay. Maybe I just suck at using it yet, will do some more testing.

RTFM that feature exists already

[Azbest]

Yep I just found it. Much better now.

[atell]

I'm demoing this right now. It sure does like bright mixes. I don't know if I'm happy about it.

edit
After some brief testing I think this plugin could benefit from an option to limit the range it works with. I like what it does to lows and mids, but I like my top end a bit subdued, and Gullfoss makes it too harsh for my taste. I have to apply an EQ after it to keep the piercing sizzle at bay. Maybe I just suck at using it yet, will do some more testing.

As pointed out by others, there are range limiters that prevent Gullfoss from acting on the top end. However, the reason why Gullfoss boosts this range is because you have audible information there that is being masked by other audible components. Namely your high piercing sizzle. It points you towards an issue with your input signal and the best strategy is to find the source of this signal component and to eliminate it instead of turning it down again. Otherwise it will come back and bite your behind, for example, if you play your music at a venue with high sound pressure level.

A good top end does not produce any sizzle or piercing noise when running through Gullfoss. So it's doing what it's designed to do. And this is usually the right explanation when reducing Brighten and lowering Boost don't work. In my experience, if you need to use the high-frequency limiter on a full mix, you have an issue with your mix.

Thanks,

Andreas

[fese]

A good top end does not produce any sizzle or piercing noise when running through Gullfoss. So it's doing what it's designed to do. And this is usually the right explanation when reducing Brighten and lowering Boost don't work. In my experience, if you need to use the high-frequency limiter on a full mix, you have an issue with your mix.

Great. I spent A hundred bucks on a plugin that tells me that my mix sucks. Don’t need no plugin for that

[Tchad Froom]

A good top end does not produce any sizzle or piercing noise when running through Gullfoss. So it's doing what it's designed to do. And this is usually the right explanation when reducing Brighten and lowering Boost don't work. In my experience, if you need to use the high-frequency limiter on a full mix, you have an issue with your mix.

Great. I spent A hundred bucks on a plugin that tells me that my mix sucks. Don’t need no plugin for that

Sounds like you should have demo'ed before buying.

[sengoku]

according to https://www.soundtheory.com/about : "Why do waterfalls sound so pleasing? To answer, one could argue that a waterfall generates close to pink noise."

This quote requires some context. It's true that the question why waterfalls sound so pleasing inspired the research that resulted in Gullfoss. But along the way, the understanding we gained went beyond the simple observation that pink noise sounds pleasant. We actually tried to understand why this is the case and arrived at more general principles. So Gullfoss does not have an internal pink noise reference.

thanks for the more detailed explanation andreas. gullfoss is an incredible plugin, excellent work.

[DJ Warmonger]

A good top end does not produce any sizzle or piercing noise when running through Gullfoss. So it's doing what it's designed to do. And this is usually the right explanation when reducing Brighten and lowering Boost don't work. In my experience, if you need to use the high-frequency limiter on a full mix, you have an issue with your mix.

Great. I spent A hundred bucks on a plugin that tells me that my mix sucks. Don’t need no plugin for that

Learning experience is worth more than any plugin

But, why not have both