AI Audio Post

Every indie filmmaker has the scene. The performance is the best take of the day — and under it there's a bus, or a refrigerator hum you swore you'd unplugged, or wind hitting the lav like a drum solo. For most of film history, that scene had two fates: live with the noise, or drag your actor into a booth months later for ADR and pray they can find the performance again. They usually can't. ADR dialogue on an indie budget almost always sounds like what it is.

AI dialogue isolation changed that math more than any other single tool in this course. Modern voice-separation models can pull spoken dialogue away from traffic, hum, rain, HVAC, and wind — not by carving out frequencies the way old noise reduction did, but by learning what a human voice is and rebuilding the signal around it. The dialogue isolation built into DaVinci Resolve and Premiere Pro, and the dedicated dialogue-isolator category beyond them, will rescue scenes you had written off. I've watched it un-cancel ADR sessions. That's money and morale saved in the same click.

The cleanup pass, in order

Audio cleanup works best as a deliberate sequence, not a panic button you hit on the worst clip. Here's the pass I run on every project once picture is locked (or close):

• 1 · Isolate dialogue. Run voice isolation on the production tracks that need it — and only those. Clean tracks should stay untouched; every pass of processing costs something.
• 2 · De-reverb. If you shot in a boomy room, reduce the reflections next. Dry dialogue is flexible dialogue — you can always add space back in the mix; you can't easily remove it later.
• 3 · A/B against the original. Every processed clip, toggled against the raw take, on decent headphones. This step is not optional, for reasons below.
• 4 · Level match. Bring all dialogue to a consistent conversational level across the film so no one rides the volume knob. AI loudness tools do this fast; your ears confirm it.
• 5 · Prep the mix. Split dialogue, music, and effects to their own organized tracks. Whether you mix it yourself or hand it off, clean stems are the difference between a mix and a mess.

For documentary and interview work, add one more tool to the kit: transcript-based cleanup. Because your edit is already tied to a transcript (Chapter 5), removing filler words — the ums, the false starts, the "sort ofs" — becomes a text edit. Used gently, it tightens an interview beautifully. Used greedily, it makes a human being sound like a text-to-speech engine. Cut the stumbles that add nothing; keep the ones that make the person a person.

Limits, and where the money goes

Now the honest part. Push these tools hard and the voice starts to thin out — a slightly underwater, swirly quality engineers call artifacts and audiences call "something's off," even when they can't name it. A voice stripped of all room tone can also feel weirdly disembodied, floating in front of the picture instead of living inside it. The fix is restraint: process at the lowest strength that solves the problem, and always compare against the original before committing. Sometimes fifty percent of the noise gone is the right answer, because a hundred percent kills the take with it.

And here's my one spending recommendation in this whole course. I've told you the free and built-in tools are enough almost everywhere — rough cuts, transcription, captions. Audio is the exception. If you're going to spend money anywhere in post, spend it on sound — better cleanup tools, or a few hours of a real dialogue editor's time on your hardest scenes. Sound is half the experience of your film and the half audiences judge subconsciously and mercilessly. A great mix makes a modest picture feel expensive. The reverse is never true.

Next chapter we move from sound rescue to picture rescue — upscaling, denoising, and restoration, and the crucial skill of knowing when footage can be saved and when it's telling you to reshoot.