Auto-mix Plugin Shootout

[kidslow]

Thanks for the additional detail and sharing the clips.

IMO, to use any of these one first has to commit to a plugin and learn it.

That seems to be true of any channel strip, or really any complex plugin for that matter, so perhaps the question is would that time be better invested elsewhere?

Do you think that a plugin like Waves Vocal Rider or Hornet Autogain would help balance the gain feeding one of the automix plugins?

[jochicago]

That seems to be true of any channel strip, or really any complex plugin for that matter, so perhaps the question is would that time be better invested elsewhere?

Do you think that a plugin like Waves Vocal Rider or Hornet Autogain would help balance the gain feeding one of the automix plugins?

I think the point is to not have to make eq and compression decisions. If you can trust that a particular plugin can mix an EL guitar fine, then you don't have to think about what to do, just apply this one plugin and let it do its thing.

Re: gain raider
Maybe. But IMO you still need to manage the settings on those gain raiders. The Waves Vocal only has 2 speed settings, so in a way that's easier but it also means you are stuck with what it does. Hornet has a ton of settings, so either use a preset or take time to manage it and have no time savings.

Perhaps something like BX_opto (LA2A style) might be easier. You set it early and make it do 5db of GR, because it's so easy to set and very transparent, you end up with a more reliable signal to feed into the mixing plugin. In any case, you'll be making some decisions and doing some mixing.

Personally, after my test, one of my biggest problems was not knowing how much compression was taking place. I had to go by ear and if it sounded fine I'd let it go, but then at the chorus I realized the dynamics were out of whack, and fixing that means going back to these plugins that have next to no settings so it's hard to solve. So running the signal by some sort of compressor/gain raider first to reduce the dynamic range transparently sounds like a good idea. Or maybe chase it with a compressor that grips tightly (say 1176), so anything that's out of whack gets tamed.

[HoRNet]

Wow lot’s of great ideas here! I wanted to create a better version of an “auto mix” plugin, even though we don’t advertise TrackShaper like one... I think a dialogue between instances is essential!

Saverio

[jochicago]

Wow lot’s of great ideas here! I wanted to create a better version of an “auto mix” plugin, even though we don’t advertise TrackShaper like one... I think a dialogue between instances is essential!

Saverio

I'm glad to hear you welcome the challenge. I think it's a big one, to try to make something that listens to all tracks. I imagine it could get very complicated quickly. The solution would have to be some elegant simple logic, but in your case with Hornet products I think you already have an engine to borrow from. For this exercise I was using Hornet 31 on the master at 30% to soften the weirdness created by these auto-mixers not listening to each other.

You could use something similar to that engine on the master bus, and use those calculations to determine which track instances need to do more or less.

Like a in a simple rock song: drums, bass, vocals, guitar
Say your master metering reads it needs more 80hz, so it has to decide if it gives it to the kick or bass. Because if it were to just raise the 80hz frequency at the master it would be raising for all instruments summing at that point including the guitar.
If it needs more 4khz it decides not to give it to vocals because of sibilance concerns so it gives it to the guitar. That kind of thinking. So maybe it makes a handful of these decisions, and for the rest just applies something similar to Hornet 31 to wrap up the song into a pleasing frequency curve.

Another experimental idea is to have bus instances. Like instead of just at the track level and then at the master bus level, you could have summing instances in the buses. For example the idea is that a drum bus instance can try to balance the drums applying an eq curve and a specialized drum bus compressor, and doing it at that level allows for a cleaner processing, before everything sums at the master.

[SeeingInMidi]

AI, machine learning is the future. Bash it all you want, at this speed, people will soon be mixing primarily with AI based plugins. AI in the audio domain is still in its infancy.

There's incredible things happening with AI based software right now. Lots of money invested into it. Eventually AI will be able to mix a whole song perfectly and capture human nuances of the mix process too. This decade is going to be really interesting in terms of AI based plugins. Seeing what other developers are doing with AI, machine learning outside of the audio domain makes me really excited.

[RobinWood]

AI

No thanks, sounds horribly boring and nothing like the art mixing is.

[SeeingInMidi]

AI

No thanks, sounds horribly boring and nothing like the art mixing is.

Reminds me of the old arguments about analog mixing. "mixing in the box is so boring and will never sound like an analog mix". These days, many top engineers are mixing all itb or using a hybrid setup. Dont be surprised to see them also using an AI hybrid setup in the future

[RobinWood]

Yes but mixing analog or mixing itb still required the knowledge, feeling and vision of mixing ^^ With that level of AI, all that is gone.

[SeeingInMidi]

Yes but mixing analog or mixing itb still required the knowledge, feeling and vision of mixing ^^ With that level of AI, all that is gone.

Fair enough, to each his own. If you want to implement AI into your workflow or not, totally up to your preference.

[kidslow]

I, for one, welcome our new AI Mixing Overlords.

[fese]

AI, machine learning is the future. Bash it all you, at this speed, people will soon be mixing primarily with AI based plugins. AI in the audio domain is still in its infancy.

There's incredible things happening with AI based software right now. Lots of money invested into it. Eventually AI will be able to mix a whole song perfectly and capture human nuances of the mix process too. This decade is going to be a really interesting in terms of AI based plugins. Seeing what other developers are doing with AI, machine learning outside of the audio domain makes me really excited.

While I agree that “AI” (ML would be the better term) will have a growing impact on the audio industry in the future, I will argue about the “mix perfectly” ability. Maybe from a technical standpoint, yes, but as we all know the technical perfect mix isn’t always the best. Sometimes it is the decision of the mix engineer to make that one instrument sound intentionally shitty, gritty and lofi so that it doesn’t sound perfect as we expect it, but it will be the sound that defines the song.
I am not sure that ML will be able to do that, as ML mostly depends on the amount and quality of the data being fed in the learning process. It’s the intent that separates us from the machines.
It is highly possible though that an ML process will be better than me at mixing, but that isn’t really that difficult a task...

...
Reminds me of the old arguments about analog mixing. "mixing in the box is so boring and will never sound like an analog mix". These days, many top engineers are mixing all itb or using a hybrid setup. Dont be surprised to see them also using an AI hybrid setup in the future

I’m pretty sure they’ll use AI if it helps them save time. Time of course is money and that is an advantage in a competitive market. Professional mix engineer’s decision how to mix is often not only a personal choice but driven by the demands of the clients who expect a mix change being done within an hour. At least that’s the impression I get when I hear interviews from those top mixers that changed to ITB mixing.

[cnt]

Nice, I'm not the only one who chose nr 5 as my favorite, the one you apparently mixed yourself, so you're doing your things right when mixing.

I agree, auto-mixing tools can help you find things per track, but will hardly make good artistic/mixing decisions.