Over the weekend, I read a fascinating op-ed in the New York Times called “How to Build Artificial Intelligence We Can Trust.” Its basic premise is that although AI is good at looking for particular patterns in data, it doesn’t factor in time, space and causality. It got me thinking about AI and music software, and whether a related argument applies.
These days, there’s no shortage of “intelligent” music software. You have smart compressors and EQs that analyze your audio and come up with optimal settings; smart channel strips that set multiple processors based on your track; and smart mastering software (both computer and cloud-based) that can analyze your mixes and create mastering settings for them.
Many of these programs use machine learning, a form of AI, to digest vast amounts of musical data to use as reference when analyzing your music. They recommend settings based on recognition of the patterns in your music. But can such software do as good a job as a musician, engineer or producer? My answer—which is not based on empirical data but simply intuition and logic—is “not really.” That’s because the software lacks an important contextual item: your creative vision.
That said, I do think such software can be quite useful. I’ve used iZotope’s Ozone mastering software quite a bit and Neutron channel strip to a lesser extent and found that they can often be convenient for getting you to a useful starting point. In fact, iZotope goes out of its way to point out that that the results from its AI-based “Assistants” are just that: starting points.
I think that’s critical. As long as you view this type of software as a tool, rather than a substitute, for the judgment of a mixing or mastering engineer, it’s fine. If the results you get don’t fit your vision, be ready to tweak them or start over. It’s similar in some ways to when you’re scrolling through presets and trying out different ones. Some work, some don’t.
I don’t have as positive a response to cloud-based mastering services such as LANDR, into which you upload your tracks and the algorithm spits out a mastered result. The problem isn’t the results per se, but the lack of ability to do much to change them. In a program like Ozone, you can open up any of its processors and adjust them as much as you want after you get the settings from the Master Assistant, but your only choices in LANDR are three levels of intensity.
So, by all means, take advantage of the power and convenience of AI-assisted software, but always keep your vision for the music at the forefront. If you’re looking to “break the mold,” you don’t want to completely rely on tools based solely on what’s come before.