The explosion of AI in the audio space is initially exciting and quite daunting at the same time.
Questions about “how it works” and “based on what” are seemingly legitimate from my perspective. More importantly, “what is an accurate description of the end result” and “how can one put it to good use” – would be first and foremost on my mind.
We set out to find some answers and pushed our entire inventory through Logic Pro’s Mastering Assistant, a newly released AI assisted plug in by Apple that is quickly earning status as top of class.
For context, we tested over 100 commercially published wav files in our audio lab. We put several decades of mixing and mastering experience into it along with several weeks of dedicated time and focus.
The first pass struck us with an initial “wow”. The AI algorithm seems to take corrective action that creates a better balance by pan shifting toward center and clearing out some mud from the middle. Previously buried instruments come to the surface and provide new opportunities to remix yet again.
From one who has been on the mixing board for a few decades, my perspective on AI in the mastering space places it squarely in the “here’s another plug in” category. It is not the panacea of all mastering hopes that will push your stuff up a few notches on its own. It is not that at all, well depending on what you are after.
The test – let's continue. Our next step was a comparison to previously mastered versions of the same tracks. Several things were immediately clear. The AI tracks offered more clarity, but that was it. They also sounded somewhat brittle and dry. It seemed like the warmth was sucked out of them – and the power in the middle mysteriously absent. When you test your A and B takes, it must be with fresh ears. I went from “wow” on day one to "uh-oh" on day two.
Once AI was done altering my track set, much more work was needed to make the tracks re-publishable. Sterilized computer music doesn’t work for me.
Restoring the set took several new passes at the EQ, our primary focus. Most tracks required a restoration of the bottom using low shelf filters. Very subtle reverbs were brought in at various places in the eq spectrum which added clarity, color and warmth without the spaciousness of a heavier verb setting. The process led to a new mastering chain that also included re-imaging, limiting and noise suppression tools, plus other post production tools arranged by order in the chain for best results.
With tracks restored and clearly refreshed, the exercise was certainly worthwhile for us and eye opening from a process perspective. Here’s what we walked away with:
For AI to adjust an audio file, it must first assess the file to identify which aspects or qualities of the audio can benefit by being altered. It needs a benchmark for comparison to derive the changes that will lead to a supposedly improved file. What is that benchmark? Whatever it is, it didn’t work for us. Despite the adjustable parameters on the tool (which often made things worse), it never brought the file home so to speak.
The Logic Pro X Mastering Assistant is a first step for us if we use it at all. It created some balance regarding generally accepted parameters required for quality stereo output. It did a fine job of cleaning out some unwanted and previously hidden middle. It brings you to a preset set of parameters that are adjustable. Nevertheless, we see it as a first step in the mastering process, yet not a necessary step if the state of the mix is highly satisfactory.
If you go with it, or tools of similar nature, the initial output in my humble opinion will be audio that is dynamically squashed, over-bright and brittle. My sense of that may be exacerbated by my preference for a warmer, vintage like sound. So much more of the work is in the mixing, where AI has no place for us. Mixing is human craft for human consumption. In comparison, technology is no match.
So, with that, if you use AI in your mastering process, it may be fine for what you do. For us, we had to double up on the effort to bring the tracks back to an acceptable level of Creative Music's studio quality standard. Does it have a place for us, yes. Is it the be all, end all? Not here. Is it useful? With the right expectations, yes.
We learn by doing, and this is what we found.
Lou @ Creative
Comments