Behind the Code: Development of Ozone 7 Advanced
We like to be innovative and scientific. We’re not focused on analog modelling, but when we do it, we like to do it really well. I remember our old CTO actually posed a question to me that was like, “If there were some aliens that had never learned anything about all of the analog hardware design, how would they be designing plug-ins?” And that’s sometimes an approach that we’d like to take, where there’s a lot of awesome things that we can do digitally that you can’t do analog.
In terms of goals for DSP development, there’s a big theme of vintage emulation. There was also just the sheer scope of the number of new algorithms that we were doing.
I mean, there’s some things that are totally in our will, we’ve done things like them before in an area where we already have some expertise and we already know how to do a really awesome job there, and then there’s other things that we have less experience with, and we know we have a really high bar for quality for ourselves, and it’s about making sure we have enough time to first learn how to do an awesome job at that, and then to actually execute.
When people use a maximizer, they’re generally using it to make their music louder and to do that transparently, so you’re not causing any clipping distortion or pumping. We’re very well known for our IRC algorithm. One thing that’s pretty unique that it does is it analyzes the properties of your signal and decides in real time how to make the best trade-offs to make it sound good.
Before Ozone 7, the most recent version of that was IRC-3, which uses some psycho-acoustic models, which are models of the human auditory system to figure out how audible different types of distortion will be, and then pick the thing that’s going to sound the best.
The basic idea with IRC-4 is rather than just adjusting the gain of your signal, it also adjusts the frequency content in real time. For example, if you have some drums and you also have some vocals at the same time, a conventional limiter, in order to prevent the drums from clipping would also cause pumping in the vocals. The IRC-4 is going to be able to reduce the drums without touching the vocals so much.
One of the things that really differentiates it from other products, other limiters, is we don’t really have a small number of bands. Typically, you’ll have three or four bands of limiting. We have many more psycho-acoustically spaced bands so we can be more smooth and selective about the frequencies it chooses to select.
Our IRC algorithm is already really good, so we knew we wanted to make it better, it was actually a challenge to figure out, “How do we make this better?” And another challenge was the spectral shaping algorithm we’re using, an effect is if, for example, the bass is really loud, it’s going to reduce the bass. If the treble frequencies are really loud, it’s going to reduce those. A challenge was making sure we don’t overdo those to make sure we’re getting the benefits of this algorithm without totally flattening everything else out.
Our approach to vintage processing is that we want the best of both worlds. The analog stuff really matters, we want to do an awesome job with that, but we also want to take advantage of the things that you can’t do in hardware that you can only do digitally.
As far as how vintage processing fits into the overall story of Ozone, I think Ozone is really well known for having clean, precise algorithms, and I think the addition of these vintage modules kind of helps round it out so you have the right tool for the job. Sometimes you want a clear, linear phase EQ, and sometimes you want a different sound.
The vintage limiter is similar to a compressor, or kind of similar to our maximizer in that it’s used to affect dynamic range, to glue things together a little bit more, and potentially to make things a little louder. It’s different from the maximizer especially in that it’s more analog based. It’s not meant to make your signals as loud as it possibly can, it’s more meant to add some vibe and character and glue things together.
We’ve modeled it, sort of loosely inspired by a Fairchild 660 or 670, which is a very old, very expensive dynamic processor that has a whole lot of tubes and is very heavy. It was tricky to make it appropriate for the mastering workflow.
One part of our emulation was actually some saturation that the tubes in the Fairchild add. There’s some tubes that are used to control the gain, they add some distortion. When we did some listening, we actually decided in a mastering context, that wasn’t so desirable, and we wanted it a little bit cleaner.
There’s a couple things that make this unique. It’s a feedback compressor, which means that it’s actually doing level detection on the output, which is a thing that’s really easy to do in analog and not so easy to do in digital. One of the things that you get from feedback compression is that the attack and release envelope are a little bit different, so that just gives it a different sound, versus our maximizer, or our existing dynamics module, it also has no look ahead, which is, again, common for analog, not common for digital.
These are all just things that give it a different sound.
Our vintage EQ is modeled on two different Pultec equalizers, the EQP-1A and the MEQ-5. So the EQP-1A is for the low and high frequencies, and the MEQ-5 is for the mid-range frequencies.
When people use this unit, one of the really unique features is that the low band can be both boosted and cut at the same time, and the manual actually tells you don’t do this, but it turns out, it sounds great, you’re able to basically boost the really low frequencies, and have cut come right after it in the low mids, and that lets you boost the lows without getting a really boomy sound, so that’s a really useful feature.
Another useful thing is this is just a more restricted workflow versus a parametric EQ, so you have slightly fewer options, and sometimes, that way of working can push you to get a different sound than you might have otherwise, or put you in a more creative headspace.
Two of the main challenges — so the first was getting an accurate frequency response, especially in the high frequencies, making sure we actually matched the behavior, and the second challenge was that the parameters all behaved smoothly, so when you turn a knob, it doesn’t click or crackle, and just generally, because of the way it’s setup, that it targets the low frequencies and the high frequencies and a few different mid frequencies, it’s helpful for sort of broad gestures of saying, “I’d like to make this a little brighter, I’d like to add some bass, I’d like to cut some mids.”
We worked with our sound designers and found that there was a low boost and cut frequency that would be really useful to have for mastering that didn’t exist on the original unit, so we actually added that in.
Obviously, people really like the sound of tape. It can glue things together nicely, it can add some saturation and alter the frequency response, so this is another creative tool that our customers will have to be able to get the sound they want.
The main challenge is just how hard tape emulation is, and it’s not so well understood. We understand how to model circuits and how tubes work. The physics of tape are a lot more complicated, so when we first set out to do this, this was a case where we had to learn a totally new set of skills, and we actually were really uncertain if we’d be able to get a good result, and we wanted to make sure we could, so I’m really proud of the result that we got. I think we learned a lot and did a really great job.
The first step is if you’re going to model something, you better model something that sounds really good. So Jonathan Wyner, a mastering engineer, and director of education at iZotope has a Studer A-810 in his studio, which he uses for mastering, and it’s in really great condition and it sounds awesome, so we started there. It’s a mastering tape machine, which means it’s actually, compared to most tape machines, pretty clean. It’s a pretty subtle effect, and one thing we wanted to do was make sure our customers would be able to get exactly the sound that they want.
So we went beyond modeling the actual machine, and provided some other controls that let you control how much of various effects you’re getting, so how much distortion you’re getting, how much of the low frequencies, or so called head bump you’re getting, things like that.
One thing that tape does is it affects the frequency response, and another thing it does is it distorts different frequencies by different amounts, so one thing we had to do was design test signals to figure out how much distortion was happening at different frequencies and make sure we were matching the behavior of tape.
Our model of frequency dependent distortion, how much frequencies distort is really accurate and matches the original machine really well, and another kind of unique thing we modeled is the phase distortion that this machine introduces. One thing that happens through interactions with the electromagnetic field is you get some amount of electromagnetic phase rotation, so we actually modeled that as part of our algorithm.
The vintage compressor is a feedback compressor with program dependent release that’s sort of inspired by a lot of classic compressors, but not directly modeled after a specific one.
In terms of what it offers as a creative tool, again, it offers control of dynamic range and a way to glue things together and get a different sound.
I’ve been describing the vintage compressor as a faithful analog emulation of a compressor that never existed, and that’s basically what it was. We started designing it by looking at a lot of classic compressors and understanding really what makes them tick, what are the qualities that they have that people really like, then trying to understand how we can get those same effects.
One particular quality about this compressor is that it’s a feedback compressor, which makes my head hurt a little bit. It means unlike a regular, or especially digital compressor, it does level detection on the output.
A typical feed forward compressor, you take the input signal, you detect the level, you just add some gain to apply to that, smooth it out, and that gives you your output.
In a feedback compressor, you’re actually detecting the level of the output of the compressor, which computers don’t really like to do. That’s something analog hardware does really well and computers don’t do so well.
So it was really a challenge to decide how to accurately emulate that.
For example, you could say, well, in order to find the output at this sample, I’ll look at the output of last sample, and that’s introducing a delay, and that actually would mess up the attack and release behavior and wouldn’t give an accurate emulation, so we’ve used some pretty cool techniques to really accurately model this thing.
Within Ozone, one thing that makes this really unique is it’s designed for a single band workflow. It’s a single band compressor and it also has no look-ahead, which gives it a particular sound.
Again, with look-ahead, you can get a little less distortion, but that’s its own sound. The workflow and the sound you can get from this look-ahead free analog style compressor is different from what you could already get in Ozone.
One case where we’ve sort of gone above and beyond and done something that would at least be difficult to do in hardware is you get a really wide array of side-chain filter options that lets you filter the signal that the compressor sees. So for example, you can high-pass out some of the low frequencies in order to reduce pumping.
As part of our process for designing this algorithm, we looked at a lot of different compressors that had side-chain filters, and we’ve sort of offered everything you could possibly want in a parametric style EQ sort of way, so instead of having to turn a knob, you can actually visualize the side-chain filter’s frequency response.
One of the main things that distinguishes this from past versions of Ozone is the focus on vintage emulation. We’ve definitely done analog emulation in the past, but we haven’t designed modules around that.
Another big difference is the sheer scope of new algorithms, especially in comparison to Ozone 6, but really in comparison to I think really any Ozone release we’ve done. There’s four new modules, there’s new IRC mode, it’s a lot of new algorithm developing.
We have a very single processing, research oriented team, but also we have some really talented sound designers, we have some great people in QA with audio engineering experience, and I think the quality of these algorithms has really benefited from all of us working together from having the ears of really skilled engineers, and skilled signal processing engineers working together.
As the manager of our research team and someone who is involved with this development effort, I’m immensely proud of everything that came out of this. There’s times you want to reach for those digital tools that Ozone has offered for some time, and I think there’s other times that you want to reach for those things with different kinds of character and those vintage emulations, so I hope people really enjoyed the opportunity to use both of those things and choose the right tool for the job.