-
AI – Your Assistant, Not Your Replacement
If you’ve ever spent a late night hunting breaths, taming sibilance, or lining up ADR takes, you already know the paradox of modern audio: our tools are more powerful than ever, yet creative time feels scarcer than ever. At NoiseWorks, our philosophy is simple: AI should be the assistant that gives engineers their evenings back—not the auteur that takes their chair.
This post lays out where AI in audio is headed, how leading tools are reshaping post-production, and where we draw the line: **humans decide, machines assist**.
What AI Is Already Great At
Dialog clean-up and enhancement.
Modern AI denoise/dereverb has crossed the “broadcast-ready” threshold. Some tools can turn rough room recordings into clean, treated-sounding VO and even regenerate clearer voice while suppressing noise/echo in near-realtime—usable in daily workflows.
Search, transcript, and structural edits.
DAWs are becoming language-aware: recent updates add AI speech-to-text and faster ADR, letting editors jump to content semantically instead of scrubbing.
Surgical repair.
Spectral repair suites remain the scalpel—isolating dialogue, painting out intrusions, de-rustling lavs, and matching ambience—with ML increasingly tackling the ugliest problems.
Generative beds, SFX & temp tracks.
Generative tools (e.g., Meta’s AudioCraft, Stability’s Stable Audio 2.5) now deliver controllable music/SFX—including on-device options via Arm—so teams can spin up temp soundscapes, alternates, and brand “sound DNA” fast.
Where This Leaves the Human Engineer
Taste, context, and narrative remain human.
AI is brilliant at what to fix; it still struggles with why something should sound a certain way. The emotional arc of a scene, the intent behind a room tone choice, or the micro-timing that sells a joke—those are editorial calls born of taste and experience.
Quality goes up when busywork goes down.
When machines handle the repeatable 60%, humans can over-deliver on the 40% that actually moves audiences. That’s the exchange we’re after.
How We Build for This Future at NoiseWorks
Our products start from a bias: put the human’s intent first, automate the rest. With DynAssist, we learned something liberating—engineers don’t want AI that “guesses the mix.” They want AI that listens like an assistant, understands the job (find breaths, shape sibilance, respect performance), and gets out of the way.
So our roadmap follows two rules:
Assist, don’t author.
Default to explainable actions with visible, tweakable parameters. Make it clear what changed and why.Make time, not decisions.
Every feature must return minutes to the editor: faster analysis passes, looped re-analysis only where it matters, session-aware presets that follow your intent across takes.
What Dax thinks about using AI
Interviewer: I’d be interested in your opinion—a lot of concern for pros in the industry right now is that AI is going to take all the good parts of their jobs.
Dax: There is a great quote about the way AI is being used a lot these days, and someone said, “Rather than taking the tedious parts out of creative people’s lives, it’s allowing tedious people to be creative,” but not actually creative; they’re just literally saying, “paint me a picture that looks like this.” It’s not really art, is it?
What I like about NoiseWorks’ approach is that they haven’t used AI for everything; they’ve only used it in certain parts, and the developers told us they tried it in different areas, and they found that for some sections an algorithm worked better than using some kind of machine-learning AI tools.
I think that’s the key to this—whenever new technology comes out, it’s always abused. Auto-Tune came onto the market, and pretty quickly we had Cher’s Believe.
That was an abuse of the technology; it wasn’t how it was originally intended to be used, and then, of course, everybody started doing it; it became overused.
I believe—I hope—that the AI side of things will follow a similar pattern: it’s all very, very new, you know, for the general consumer.
We now have access to tools we wouldn’t have dreamed of.
I feel like we are going to get to the point where people are like, “Okay, right, we had fun using it for everything—now let’s work out where it’s best applied: what should go back to humans, where a straightforward algorithm is enough, and which things we should use AI for.”
And the beauty of DynAssist is that one of the most tedious parts of our job is being removed.
The Line We Won’t Cross
We build audio assistants that never tire and never miss a breath. We won’t build black boxes that make aesthetic choices for you. The future of audio belongs to teams that use AI to protect the human parts—taste, storytelling, passion, and creativity.
