Workflow guide
Narration that sounds recorded, not captured
Voiceover carries explainer, tutorial, and documentary video — viewers will tolerate modest visuals under a confident, clean read. Most bad narration is a room problem and a pacing problem, not a voice problem. This guide takes you from recording chain to a mix where every word lands.
From the Mellius product team · Updated September 28, 2026
Conspiracy theorists, you may hate this
AI Director
›
Just type what you need
- ›Remove all silent parts
- ›Add bold kinetic captions
- ›Only have Bill Ackman speaking
- ›Cut to highlights only
- ›Add cinematic transitions
- ›Match cuts to the beat
The short answer
Record in a small soft-furnished room with a cardioid mic six to eight inches away, capturing at minus 12 to minus 6 dB peaks, then edit breaths and mistakes before placing narration on its own track. Duck background music 6 to 10 dB under the voice, keep sentences paced to the picture, and export with loudness normalized.
Clean signal
No room echo, hiss, plosives, or clipping competing with the words.
Paced to picture
Sentences land on the visuals they describe with room to breathe.
Sits above music
Narration stays effortlessly intelligible with the bed tucked underneath.
Treat the room and set the chain
Record in the smallest soft-furnished space available — a closet of clothes outperforms an empty living room — with the cardioid mic six to eight inches off-axis from your mouth and a pop filter in between. Set gain so peaks land between −12 and −6 dB, leaving headroom for emphatic lines, and record a few seconds of room tone for patching later. Kill noise at the source: fridge, fan, notifications, and hard reflective surfaces near the mic.
Monitor on closed-back headphones while recording so mouth clicks, plosives, and chair creaks get caught in the moment rather than the mix. Record each section as separate takes with two seconds of silence between them, which makes comping the best read dramatically faster. Common mistake: recording across the room from a laptop mic and hoping software rescue will fix it — cleanup helps, but distance bakes echo in permanently.
- Mic 6–8 inches away, slightly off-axis, with a pop filter.
- Peaks at −12 to −6 dB; capture room tone for patching.
- Silence appliances, notifications, and reflective clutter first.
- Monitor on closed-back headphones and re-take on the spot.
Direct the read, then edit for pace
Read standing if energy lags and sitting if it rushes, marking the script where each line meets its visual so pacing serves the picture. Punch in on flubbed sentences rather than restarting whole paragraphs — matching tone across one sentence is far easier than across five. Leave deliberate pauses between sections; you can tighten them in the edit, but you cannot invent breathing room that was never recorded.
Comp the best takes onto one narration track, then de-breath lightly rather than stripping every inhale, which makes voices sound robotic. Remove mouth clicks, tighten gaps to a natural rhythm, and apply gentle de-essing only where sibilance bites on small speakers. In Mellius, narration sits as an ordinary editable timeline track, so reordering a section or re-pacing a pause is a conversational request with undo behind it.
- Mark script-to-visual sync points before recording.
- Punch in per sentence; keep pauses generous at capture.
- Reduce breaths instead of deleting them entirely.
- Comp to one continuous narration track before mixing.
Mix narration over music and export
Duck the music bed 6 to 10 dB under narration with sidechain compression or manual keyframes, riding it back up in pauses so the track breathes instead of pumping. High-pass the voice gently around 80 to 100 Hz to clear rumble, add a small presence lift only if the read sounds dull on phones, and keep processing subtle — heavy compression on an uneven read exaggerates every flaw. Balance once on monitors, then confirm intelligibility on a phone at low volume.
Export with loudness normalized to your platform's target — roughly −14 LUFS for YouTube and social — and keep true peaks below −1 dBTP so encoding never clips. Archive the raw takes alongside the finished narration track so pickups match the original chain. A free start in a web-based editor means you can record, place, and revise voiceover without installing a studio's worth of software first.
Compare the workflows
| Setup | Best fit | Trade-off |
|---|---|---|
| Built-in laptop or phone mic | Scratch reads and quick drafts | Room echo and noise baked in |
| USB mic plus treated room | Most creator and course narration | Needs setup discipline every session |
| Studio or hired voice artist | Brand-critical and broadcast delivery | Highest cost and scheduling overhead |
A practical sequence
Put it into practice
- 01
Set the chain
Treat a small room, place a cardioid mic close with a pop filter, and peak at −12 to −6 dB.
- 02
Record and comp
Punch in per sentence against script-to-visual marks, then comp the best read to one track.
- 03
Duck, balance, export
Tuck music 6–10 dB under voice, check on a phone, and export loudness-normalized.
Frequently asked questions
How do I remove room echo from a voiceover?
Prevention beats repair: record close to the mic in a small soft-furnished space so reflections never dominate. Light de-reverb and gating can tame mild boxiness, but heavy processing sounds hollow and fatiguing. If echo is obvious on phone speakers, re-record the section rather than fighting it.
Should I record voiceover before or after editing video?
Record to a locked or near-locked cut so every line meets its visual without stretching or cramming. A scratch read early helps time the edit, but the final performance should follow picture lock. Re-record pickups against the finished timeline rather than bending the timeline around old audio.
How much should music duck under narration?
Start with the bed 6 to 10 dB below the voice and adjust by ear on small speakers. Ride the music back up in narration pauses so the track breathes, using sidechain compression or manual keyframes. If any word needs effort to catch, the bed is still too hot.
Continue reading
Primary sources
Product capabilities change. These official references support the comparisons above.
Describe the edit. Keep the timeline.
Mellius turns conversational direction into cuts, captions, graphics, motion, color, and sound you can still inspect and change.
Start editing free