Workflow guide
Clean dialogue without the underwater robot voice
Noise removal fails when one aggressive pass strips hiss, hum, and room tone together and takes the voice with them. Professionals stack gentle fixes, each targeting one noise type. This guide walks that stack from recording triage to final polish.
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
Identify whether the noise is hum, hiss, or room tone, apply a high-pass near 80 Hz, remove hum surgically, run voice denoise in two light passes rather than one heavy pass, then add faint room tone so edits and gaps sound continuous.
Voice stays natural
Consonants and breath sound human, with no warbling or metallic artifacts.
Noise is unnoticeable
Hiss and hum sit below perception at normal listening volume.
Gaps sound continuous
Pauses and cut points share one smooth ambience instead of digital silence.
Diagnose the noise before touching a slider
Different noises need different tools. Electrical hum lives at 50 or 60 Hz plus harmonics and yields to a narrow notch or de-hum processor. Broadband hiss from preamps and fans needs a voice denoise or spectral tool. Intermittent sounds — traffic, keyboard clicks, air conditioning cycles — respond better to editing and room-tone patching than to global processing.
Start every cleanup with a high-pass filter near 80 Hz for male voices and 100 Hz for many female voices to remove rumble that eats headroom. Then sample a noise-only section for learning-based reducers rather than processing blindly. In Mellius you can ask for dialogue cleanup conversationally and review the result against the original on the timeline, keeping the change reversible.
- Hum: notch 50/60 Hz first, then harmonics.
- Hiss: light spectral or voice denoise, never maxed out.
- Intermittent sounds: cut, patch, or re-record beats.
- Always high-pass rumble before denoising.
Stack light passes and restore ambience
Two gentle denoise passes beat one aggressive one. Run the first pass until hiss just becomes acceptable, render or commit, then run a second lighter pass only if needed — artifacts compound fast past 50–60% reduction on most tools. De-reverb deserves the same caution: pull room echo back slightly rather than chasing a dead studio sound that makes voices thin.
Finish by re-adding life: lay a low bed of clean room tone captured from the shoot under all dialogue so pauses and cut points share one continuous space. Common pitfall is total digital silence between lines, which reads as broken audio on headphones. Level dialogue consistently before denoising so quiet lines do not get processed harder than loud ones.
- Two light passes instead of one heavy reduction.
- Keep denoise under ~60% per pass to avoid artifacts.
- Bed clean room tone under every dialogue gap.
- Level-match clips before processing, not after.
A practical sequence
Put it into practice
- 01
Isolate and filter
Identify hum, hiss, or intermittent noise, then high-pass rumble near 80–100 Hz.
- 02
Reduce in light layers
Notch hum, then run voice denoise gently in two passes with artifact checks.
- 03
Rebuild the bed
Add faint room tone under gaps and verify on headphones at normal volume.
Frequently asked questions
Why does noise removal make voices sound robotic?
Over-processing: the reducer is removing voice frequencies along with noise, creating warbling artifacts. Back the reduction under 50–60%, use two lighter passes instead of one heavy pass, and process level-matched audio. If artifacts persist, accept slightly more hiss — listeners forgive hiss faster than robot voice.
Should I remove echo or reverb from dialogue?
Reduce it lightly rather than eliminating it. Heavy de-reverb thins voices and sounds unnatural fast. A small amount of room character helps dialogue sit in the scene. For very echoey rooms, combine mild de-reverb with tighter mic technique or ADR on the worst lines.
Is it better to fix noise while recording or in post?
At the source, always. Move mics closer, turn off fans and fridges, and record 30 seconds of room tone for patching. Post-production cleanup rescues usable takes but costs time and voice quality. The best edit starts with the quietest possible recording.
Continue reading
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