For years, I have more or less dismissed Ozone’s automatic mastering.
Every now and then I have tried it, and my reaction has usually been more or less the same:
I don’t like what this does to the mix.
But one particular experience made me think about why I react that way.
The one track I couldn’t get right
I was mastering an album and there was one track I just couldn’t get right.
The rest of the album was working well, but this particular song kept fighting me. I had already made several attempts at it, probably three different revisions, and I still wasn’t happy.
At that point I wasn’t looking for something to master the song for me.
I was looking for ideas.
I wanted some kind of outside input that might point me in a direction I hadn’t considered yet.
So, partly out of desperation, I tried a few automated AI mastering services, including Ozone.
The idea was basically:
What would an automated system do with this track that I haven’t thought of?
Maybe it would make some strange EQ move that would give me an idea. Maybe it would approach the dynamics differently. Maybe it would reveal something about the mix that I had become too close to hear.
When I heard the Ozone master, my immediate reaction was that I really didn’t like it.
It had changed the mix too much. It no longer felt like the song I had been given to master.
So I didn’t use it.
I made my own master, eventually got the track where I wanted it, sent the album to the client and everyone was happy.
And that was basically the end of the story.
Or so I thought.
Then I heard it again without knowing what it was
A month or two later — I don’t remember exactly how long — I came across that Ozone master again while working with PER, my blind listening application that I released through Patreon.
PER is designed for comparing different versions of the same audio without knowing which one you’re listening to, with loudness matching so that level doesn’t influence the result.
I included the old Ozone master in a blind test against the different revisions of my own master.
By then I was no longer sitting there comparing both versions against the original mix.
I wasn’t thinking about the other songs on an album.
I wasn’t thinking about the client.
I wasn’t thinking about what I had wanted to preserve from the mix.
I was simply listening to two finished masters and asking:
Which one do I prefer?
And the Ozone master won.
That was interesting.
Because when I originally heard exactly the same master, my reaction had essentially been:
This sounds terrible.
Now, without knowing what I was listening to and with some distance from the project, I preferred it.
So what had changed?
Not the master.
My relationship to the mix had changed.
Maybe Ozone wasn’t wrong
That experience made me reconsider why I have never liked working with Ozone.
Previously my conclusion had basically been:
“I don’t like Ozone masters.”
But that obviously wasn’t the whole story.
When I listened without knowing which version was which, and without constantly referencing the original mix, I actually preferred what Ozone had done.
So perhaps the problem wasn’t that Ozone had made a bad-sounding master.
Perhaps the problem was that it had made the wrong master.
And I think that distinction is extremely important.
What exactly is a “starting point”?
After analyzing your material, Ozone’s Master Assistant gives you what iZotope describes as a starting point.
This is where I think the fundamental problem begins.
To me, the starting point of a master is the mix.
Maybe I increase the level so I can hear it closer to its intended final loudness. Maybe there is some compression and limiting involved.
But fundamentally, I want to start with the mix.
Then I listen.
What does this mix need?
Maybe it needs EQ.
Maybe the low end needs some control.
Maybe something is happening dynamically that I want to address.
Maybe there is a stereo problem.
Maybe it needs to be considerably louder.
Or maybe the mix is excellent and needs almost nothing.
That is mastering to me.
Ozone approaches the problem from almost the opposite direction.
It analyzes the track and uses what it has learned from mastered music to move things like tonal balance, dynamics, and stereo image toward targets.
And then that becomes the starting point.
I think this is backwards.
The mix should be zero
Imagine another way of designing exactly the same system.
Let Ozone analyze the track.
Let it compare the mix with all the professionally mastered music it has learned from.
Let it determine that the low end is unusual, the top end is darker than average, the dynamics are greater than normal, or the stereo image is narrower.
That’s potentially very useful information.
But don’t change anything yet.
Let the mix be zero.
Then give me a control.
At 0%, I hear the original mix.
At 10%, Ozone moves the tonal balance slightly toward what its analysis suggests.
At 30%, a little more.
At 100%, I get Ozone’s full recommendation.
Do the same for dynamics.
Do the same for stereo.
Do the same wherever it makes sense.
Now I am still mastering the song.
Ozone is assisting me.
There is a huge difference between those two things.
Different doesn’t mean wrong
This is the central problem I have with using statistical targets for mastering.
Let’s say a mix has considerably more bass than most commercially released tracks in the same genre.
That could absolutely be a problem.
But it could also be what makes the track great.
Maybe the artist deliberately made the entire record incredibly dark.
Maybe the stereo image is unusually narrow.
Maybe the midrange is aggressive.
Maybe the drums are much more dynamic than what is normal for the genre.
Those things can be mistakes.
But they can also be artistic decisions.
Deviation from the average does not automatically mean that something needs correcting.
And when I receive a mix from a client, I have to assume that the mix contains intentions.
I can’t begin by assuming that everything unusual about it is a problem.
Ozone removes the individuality first
This is what bothers me about the Master Assistant workflow.
Ozone makes a collection of decisions first.
It moves the track toward what it thinks a finished master should be.
Then it tells me:
Here’s your starting point.
But from my perspective, it has potentially already removed some of the things that made the mix individual.
And now, if I don’t like the result, I’m supposed to work backwards.
Perhaps I think the vocal no longer sits correctly.
Why?
Is it the static EQ?
The dynamic EQ?
The compression?
Something happening in the stereo processing?
The limiting?
Or an interaction between several of those things?
Instead of mastering the mix, I’m suddenly troubleshooting Ozone’s master.
I’m trying to figure out which decisions somebody else made that I need to undo.
That makes very little sense to me.
If I want to preserve more of the original mix, why shouldn’t preserving the original mix be the starting point?
The strange thing is that Ozone’s master may sound better
And this takes me back to my blind test.
Because Ozone’s master didn’t necessarily sound bad.
Quite the opposite.
When I had enough distance from the project, I preferred it.
That actually makes the problem more interesting.
It would be easy to dismiss Ozone if it simply produced bad masters.
But that isn’t what happened.
It produced something I thought sounded very good when judged purely as a finished piece of audio.
It just wasn’t what I thought I should have delivered to that client.
Those are two completely different questions:
Which version sounds best to me?
and:
What is the right master for this mix and this client?
Professional mastering has to care about the second question.
Your client already made a lot of decisions
When somebody sends me a mix, I don’t consider it raw material for me to redesign.
The mix itself is communication.
The amount of bass tells me something.
The vocal level tells me something.
The dynamics tell me something.
The width tells me something.
Even things that I initially think are wrong tell me something.
Of course I make changes.
Sometimes quite large ones.
But I want there to be a reason for every one of them.
If a client sends me a mix they love and I return something that they barely recognize, it doesn’t matter very much if I can demonstrate that my version has a more conventional tonal balance.
If their reaction is:
“What happened to my mix?”
then I probably haven’t done a very good mastering job.
In my experience, that is one of the easiest ways to get a master rejected.
For mastering your own music, this is completely different
There is an important exception to all of this.
If you’re mastering your own music, use Ozone however you want.
Run Master Assistant.
If it transforms the track and your reaction is:
“Yes. That’s it.”
then it worked.
It’s your music.
You’re the person making the creative decision.
Maybe Ozone takes the track somewhere you hadn’t considered. Maybe it pushes the low end, changes the stereo image and completely alters the tonal balance — and you love it.
Great.
In that situation Ozone can almost become part of the creative process.
I can also completely understand why someone without much mastering experience can be impressed by the result.
You have a mix. You run an analysis. A few seconds later it is louder, more controlled and perhaps much closer to what you associate with a finished commercial record.
There is real value in that.
My problem is specifically with using that philosophy for professional mastering of somebody else’s music.
Ozone can still be used professionally
Obviously, this doesn’t mean that you can’t load Ozone in a professional mastering session.
You can use its individual processors.
You can use a limiter.
You can use EQ.
You can use whatever module happens to solve a problem you have.
A tool is a tool.
What I don’t think works for professional mastering is the basic Master Assistant philosophy of first creating its idea of the master and then asking the engineer to modify that result.
That is the part I think starts from the wrong end.
Give me the information, not the answer
The frustrating thing is that Ozone probably has access to exactly the information I would like to have.
Tell me:
This mix has considerably more energy below 80 Hz than most comparable masters.
The upper mids are unusually restrained.
The track is more dynamic than similar releases.
The stereo image behaves differently from the reference material.
Excellent.
Now I know something I didn’t know before.
But then let me listen and decide whether any of those things are actually problems.
And if I decide they are, let me gradually move the mix toward the suggested target.
That would be an incredibly useful mastering assistant.
It would actually assist.
The starting point already exists
That blind test taught me something useful.
I had spent years assuming I didn’t like Ozone because its masters weren’t very good.
That wasn’t necessarily true.
The Ozone master I initially rejected was actually the version I preferred when I later listened without context.
What I disliked was what had happened to the mix.
And for professional mastering, context matters.
The goal isn’t simply to create the most impressive piece of audio when heard in isolation.
The job is to take somebody else’s mix and help it become the best version of what it already is.
Sometimes that requires significant changes.
Sometimes almost none.
But those changes should be decisions, not assumptions.
So I don’t think Ozone needs a better starting point.
I think it needs to recognize that the starting point already exists.
It’s the mix.