What is ADR? How to re-record and replace a line of dialogue

What does ADR stand for in film?

ADR stands for automated dialogue replacement, also called automatic dialogue replacement, looping or post-sync. The ADR meaning is the same in every case: an actor watches a scene and says a line again so it can replace the one recorded on set.

The idea is almost as old as film sound. According to a history in CineMontage, the Motion Picture Editors Guild's journal, dialogue was being post-synchronised by 1928, and by 1935 "looping" was used throughout Hollywood: a short piece of film was spliced into a loop and played over and over while the actor matched it. Electronic looping systems arrived in 1968, and the name "Automated Dialogue Replacement" first appeared in print in 1973. ADR in film today still works the same way, only with a timeline instead of a loop.

When do you need to re-record dialogue?

When the line itself cannot be rescued another way. ADR is the last fix, not the first, because a re-recorded line is hard to make sound as real as the original. Try these in order:

The problem	Try first	When ADR is the answer
A noise over a line: a plane, a door, a cough	Voice clean-up, or the same line from another take	The noise sits on the words themselves
A fluffed or mispronounced word	A cleaner reading from another take	Every take has the slip
The script changed after the shoot	Cover the line with a cutaway and a voice-over	The speaker is on screen saying it
A line was off-microphone or clipped	Another take, or the second microphone	No usable recording exists
A clean version without swearing	A bleep or a cut	The line has to play naturally

Removing background noise covers the first fix, and choosing the best take the second.

How does ADR work?

ADR works in three steps: record a new take of the line, line its words up with the lips on screen, then make it sound like the original room. The first is a performance; the other two are matching.

The lining up is the hard part. Viewers notice when sound runs ahead of the lips far sooner than when it lags: ITU-R BT.1359, the broadcast recommendation on sound and picture timing, puts the point where people start to notice at about 45 milliseconds early or 125 milliseconds late. A word that starts a frame off reads as fine. A sentence that drifts by a few frames reads as dubbed.

Then the take has to match: the same loudness, the same tone (a close studio microphone sounds brighter and fuller than a lapel microphone across a room), the same echo, and the same background hum under it. Leave any of those out and the line jumps out of the scene.

How do you record a good ADR take?

The better the take, the less fitting it needs:

Use the same kind of microphone. If the scene was recorded on a lapel microphone, record the new line on one, at the same distance from the mouth.
Match the energy, not just the words. A line shouted across a field cannot be replaced by one said quietly in a booth. Stand, move, raise your voice as you did on set.
Watch and listen first. Play the original line on headphones a few times, then say it straight after, keeping the rhythm and the stresses.
Record several takes. Speed and emphasis change from take to take; one usually fits better than the rest.
Record in a quiet, dead room. A dry take can be given the original room's echo; an echo baked into the take cannot be taken out.
Record a little room tone on set, if you can. What is room tone explains why it saves a line like this.

A man in headphones and glasses speaks a line into a studio microphone with a pop filter, mid-word. — A good ADR take is said standing, at the original's energy, with the original line playing on headphones first.Photo by ANTONI SHKRABA production on Pexels

How do you replace a line of dialogue in Sharpy?

You ask the agent. Its fit_line tool takes a new take of the line, whether it is your re-recording, another reading by the speaker or a line said in Sharpy's own voice, and fits it to the original's words. Inside an edit, the same fit is edit_clip's ADR action, and it is one undoable step.

Say which line. Tell the agent which line in the edit the new take replaces.
Give it the new take. A re-recording, or the line typed out for Sharpy's voice to say.
Say how many lost words you allow. None, unless you say otherwise. A fit that would lose more is refused and its file removed, so a damaged line never lands on your track.
Let it fit. Each word starts where the original's did and the pauses absorb the difference. Sharpy only shortens speech where the original gave it less time, and never slows it down.
Check the report. A word missing from the take, or said twice, is found and reported rather than mistaken for a shift.
Listen in place. The new line replaces the old one on the track; undo puts the original back.

Sharpy aims to start every word within a frame of the original's. On a test sentence 3.9 seconds long, Sharpy's voice read it at three paces, 5.4, 6.2 and 6.5 seconds long; all three fitted into the 3.9 seconds and read back word for word. Each fit took about 23 seconds on the base Mac mini M4 with 16 GB.

How does Sharpy make the new line sound like the original room?

It measures the original, then gives the take the same sound, without moving a word:

Loudness. The take is brought to the original line's level. On the test sentence it was raised 2.2 to 2.5 dB.
Tone. The take's tone is moved towards the original's, so a booth microphone stops sounding brighter than the scene.
Echo. Sharpy measures how long the original room rings after each word, from the line itself, and gives a dry take the same room. On a test recording whose room rang for 0.57 seconds, Sharpy's voice rang for 0.17 seconds bare and 0.54 seconds once fitted, and the match is printed beside the result. You can turn it off with decay: false.
Room tone. The background from the pauses around the line is laid under the whole new line. In the test, the room around the line measured −39.4 dBFS, and so did the gaps in the fitted take.

Can ADR use a synthetic voice?

Yes, with care. Sharpy can say a line in any of its 54 voices, and that voice is always declared: a piece that carries it says so in its description and in the file's Content Credentials. Use it for a narrator's line or a placeholder, not to put words in a real person's mouth. AI voice-over for videos covers when a synthetic voice fits.

For the speaker's own voice there is overdub: retype a word and Sharpy says it in the speaker's voice and matches the mouth. It does this only with the speaker's recorded permission, and the export carries Content Credentials and an invisible watermark marking it as altered. The EU AI Act's Article 50 asks for synthetic speech to be marked in a machine-readable way from 2 August 2026, and Sharpy marks every piece that carries its voice.

Frequently asked questions

Is ADR the same as dubbing?

No. ADR replaces a line in the same language with the same speaker or role; dubbing replaces the speech with another language. The fitting is similar, which is why AI dubbing uses the same timing rules.

Why does ADR sometimes sound fake?

Because one of the matches was skipped: the timing drifts, the tone is brighter, the room's echo is missing or the background drops to silence under the line. Each is measurable, so each can be fixed.

Can I replace a single word?

Yes. Re-record the sentence around it so the rhythm is natural, then fit it. Fitting a single word on its own rarely sounds right.

Does fitting a line slow my voice down?

No. Sharpy only shortens speech where the original gave it less time, and lets the pauses take up any difference.

Does my recording leave my Mac?

No. The fitting, the measuring and the render all run on your Mac.

Sources

Checked on 18 October 2026.