Making a replaced line sound like it belongs

The problem

You re-record one line. Or you use a clean take of a sentence from a different part of the interview. Or you have narration recorded in your bedroom placed over footage from a warehouse.

It sounds wrong, immediately and obviously, even though the voice is the same person and the level matches. What has changed is the space — the pattern of reflections that tells the ear how big the room is and how far away the speaker was.

What the ear is detecting

Three things, in order of how much they matter:

Reverberation. How long reflections take to decay. A bathroom decays slowly, a bedroom with a bed and curtains decays quickly, outdoors barely at all. A dry line in a reverberant scene sounds like it was pasted on, because it was.

The direct-to-reflected ratio. How close the microphone was. A close-miked line has a high proportion of direct sound and reads as intimate; a line recorded across a room reads as distant even at the same volume.

The noise floor. Every room has one, with its own character. A perfectly silent line in a scene with a fan running is a hole, the same hole the room tone lesson described.

Fixing it, in order

1. Match the noise floor first. Lay the scene's room tone under the replaced line. Half the problem usually disappears here, because the ear stops noticing the silence and stops interrogating the rest.

2. Match the EQ. A room's resonances colour the voice. Put the original and the replacement side by side, listen alternately, and use the EQ moves from earlier in this block to bring the replacement toward the original. Usually this is a small cut or boost in the 200–600 Hz region.

3. Add reverb to match, never to impress. Use a convolution reverb if you have one — it applies a measured room's actual response rather than an approximation. Resolve's Fairlight has Reverb; Reaper includes ReaVerbate and the convolution plug-in ReaVerb, both free.

Set it very low. The amount of reverb that makes a line match a room is far less than the amount that sounds like reverb. Start at 5% wet and raise until the line sits, then back off.

4. Match the perceived distance. If the original was further away, roll off some high frequencies as well as adding reverb — air absorbs high frequencies over distance, and a distant sound is duller as well as more reverberant.

Recording your own impulse response

If you can get back to the location, this is the professional method and it is free.

A convolution reverb needs an impulse response: a recording of how the room responds to a very short sound. Stand where the subject stood, put the recorder where the camera was, and clap once, sharply. Record the clap and its full decay into silence.

That recording, loaded into a convolution reverb, applies that exact room to anything you put through it. It works startlingly well, and it costs one clap.

Recording the room tone and an impulse response at the end of every shoot takes ninety seconds and solves a class of problem you cannot otherwise solve.

Going the other way: de-reverb

The opposite problem — a line with too much room, that needs to sit in a drier scene — is much harder, because the reflections are mixed into the same signal as the voice.

De-reverberation tools exist and have improved considerably. They work by estimating the reverberant component and subtracting it, and they fail in a specific way: pushed hard, the voice develops a thin, gated quality with the tails of words cut short, and consonants sound processed.

The rule is the same as for noise reduction: if you can hear the de-reverb, there is too much of it. A partial improvement that leaves some room is better than a clean-sounding artefact, because audiences forgive a reverberant recording and do not forgive a voice that sounds like a machine.

Free option: Resolve's Fairlight has De-Reverb on some tiers; check yours. Otherwise the options are mostly paid, and the honest advice is to record closer next time.

Narration over location footage

A special case worth naming because it is so common, and the answer is counter-intuitive: do not match it. Narration is conventionally understood as coming from outside the scene, so it belongs dry, close and clean. Adding the warehouse's reverb to a voice-over makes it sound like somebody standing in the warehouse, which is not what a narrator is.

Match the room only for speech that is meant to be in the scene.

Today

Record a clap in the room you are in, sitting still with the recorder a few metres away. That file is an impulse response, and you have just started a library.