Voice cleanup checklist: room tone, de-ess, noise gate, in that order
Six passes that turn a phone or lav recording into a usable voice track, each with the ffmpeg filter and a starting parameter to adjust by ear, in the order that keeps the gate off word endings.
Run in this order on the voice track only, never on the mix: 1. ROOM TONE. Record 10 s of silence in the room at the start of every session. Use it to fill gaps under cuts instead of digital silence, and as the noise profile for step 3. 2. HIGH-PASS. Cut the rumble below the voice: highpass=f=80 (f=100 for a high voice, f=60 for a deep one). 3. NOISE REDUCTION, gentle. afftdn=nf=-25:nr=12 (nf = the noise floor in dB; nr = how many dB to remove; above about 15 the voice starts to warble). 4. DE-ESS. deesser=i=0.5:f=0.5 (i = intensity 0-1; raise until the s sounds soften, stop before a lisp). 5. GATE, last and light. agate=threshold=0.02:ratio=2:attack=20:release=250 (threshold in linear amplitude; ratio 2 is a soft gate, not a hard cut; release 250 ms keeps word endings). 6. LEVEL. loudnorm=I=-16:TP=-1.5:LRA=11. ffmpeg, all passes in one chain: ffmpeg -i voice.wav -af "highpass=f=80,afftdn=nf=-25:nr=12,deesser=i=0.5:f=0.5,agate=threshold=0.02:ratio=2:attack=20:release=250,loudnorm=I=-16:TP=-1.5:LRA=11" voice.clean.wav
Paste into your assistant as the first message of a new chat; fill the braces first.
Steps
- Listen to the raw track once with headphones and write down what is wrong (hum, room, s sounds, breaths). Apply only the passes for what you heard.
- Run the chain, then A/B a 10-second stretch with a quiet word ending against the original. If the ending is clipped, raise the gate's release or lower its ratio before touching anything else.
- Keep the raw file. Every pass is destructive; the chain is cheap to re-run.
Tips
- The gate goes last because the noise reduction and the high-pass already lowered the floor; a gate set on the raw floor cuts breaths and tails you wanted.
- Two gentle noise-reduction passes beat one hard one; a voice under heavy afftdn sounds underwater.
- Shot on a phone with the radio on? The music is not noise and no filter above removes it: separate stems first (the Demucs entry) and clean the vocals stem.
| works with | ffmpeg: A terminal command you run yourself. · CapCut / any editor: A manual recipe for any editor. · By hand: No tool needed: a checklist or a technique. |
| cost | Free. No API, no key. |
| status | Unverified: written from the technique, not yet run end to end by us. |
| license | CC BY 4.0 for the text here; skills are MIT. |
| updated | 2026-11-01 |
| for agents | voice-cleanup-checklist-room-tone-deess-gate.md |
Related
Level every chapter to -16 LUFS with ffmpeg loudnorm
One ffmpeg line that brings a file to -16 LUFS integrated and -1.5 dBTP true peak without touching the video, the two-pass version for exact results, and the timestamp trap on upload.
Strip other people's music before upload with Demucs and ffmpeg
Car radio, shop speakers and festival stages trigger Content ID. Split the soundtrack into stems, find the music on drums plus bass, keep the voice, replace the music, rebuild audio only.
Live music in frame: one continuous audio span, picture cuts on top
When a band or a busker is in the shot, keep one unbroken stretch of that clip's audio under the whole scene and cut only the picture; cut the audio per shot and the beat restarts.
Score with a hole, not ducking, where something plays live
Cut the score to silence under live music, a laugh or a knock instead of ducking it; ducking is audible, a hole is not. The rule, the measured levels and the ffmpeg line.
Before you post a generated video: disclosure and rights checklist
Six checks for AI-generated clips before they go on TikTok, Reels, Shorts or a client's site: labels, likeness, music, logos, model terms and provenance.
Beat-sync cutting: cut on the beat, not on every beat
A rule and an agent prompt for cutting to music from a list of beat timestamps: cuts land on beats, phrases get room to breathe, and the best shot is held through the hook.