Denoising dialogue: how much is too much, and what to do before you reach for a plugin

Denoisers have become very good, which is exactly the problem. It is now easy to remove so much noise that the dialogue sounds like it was recorded in a swimming pool, and to do it without noticing until the mix stage.

Step zero: pick the right track

Before touching a plugin, audition every microphone on the take. A lav under a jacket may be cleaner than a boom fighting an air conditioner, or the reverse. On a poly WAV that means splitting to named mono stems first, so you can actually compare boom against each lav rather than a pre summed mix.

Step one: edit, do not process

• Cut out noise between lines instead of gating the whole track.
• Fill the gaps with matching room tone so the bed stays continuous.
• Replace a single ruined word from an alternate take before considering ADR.
• Fix clicks and mouth noise by hand, they survive denoising anyway.

Step two: gentle broadband reduction

Learn the noise profile from a clean section of room tone, then apply the smallest amount that makes the problem recede. Six to eight decibels of reduction is usually plenty for dialogue. If you need twenty, the take needs a different solution.

The artefacts that betray you

• Metallic ringing around sibilants, often called birdies.
• Consonants losing their edge so the line sounds mumbled.
• Room tone pumping in and out around each phrase.
• A hollow, phasey character on wide stereo material.

Step three: put a bed back

Perfectly silent dialogue sounds unnatural next to production sound. After cleaning, lay continuous room tone under the scene at a low level. The ear reads the result as a quiet room rather than a processed track.

Split your poly takes into named stems first, so denoising happens on the best microphone rather than on a compromise. Everything runs locally in the browser here.