Dialogue editing

Dialogue is the backbone of almost every narrative film, and audiences will forgive almost anything except dialogue they cannot understand or that draws attention to itself. The dialogue editor's job is to take the production sound attached to the picture edit and turn it into a seamless, intelligible, consistent dialogue track: choosing the best microphone and take for every line, removing noises, smoothing backgrounds across cuts, and preparing the tracks so the re-recording mixer can work quickly. Done well, dialogue editing is invisible. The audience simply hears the characters speaking in a believable space.

What the dialogue editor receives

After the conform described in lesson 5, the dialogue editor has a session in which every clip of the picture edit is linked to its original multitrack production file. Under each clip of the mix track the editor can reach the boom, each lavalier, any plant microphones, and handles before and after each cut. The editor also has the sound reports, the script, the spotting notes and the director's comments. The first pass is listening: watching the film with dialogue only, noting problems such as noise, distortion, off-microphone lines, clothing rustle, lip smacks, overlaps and background jumps.

Dialogue editors work in a digital audio workstation with the reference picture locked to timecode. They use headphones and, for final judgments, calibrated loudspeakers at a consistent level, so that noise and tonal changes are judged as the mixer will hear them.

Choosing microphones and takes

For each line, the editor chooses which microphone to use. Booms usually give more natural perspective and match the shot; lavaliers give closer, cleaner sound when the boom was too far away or the location was noisy. A good rule is to use the boom where it is clean and close enough, and to switch to the lavalier when necessary, but not to switch back and forth within a sentence, because the change in tone and perspective is audible. Where two microphones are blended, the editor checks phase to avoid the comb filtering discussed in lesson 2.

Sometimes the best solution is a different take. If a word is lost under a truck in the chosen take, the same word from another take, with matching performance and timing, may fit. These alternative takes, called alts, are a dialogue editor's main tool before resorting to ADR. The editor must match not only the words but the emotional tone, the pace and the mouth movements, so that the replaced word is invisible against the picture. Small timing adjustments of a few frames are possible; large ones show as lip sync errors.

Splitting tracks and checkerboarding

A raw edit puts all dialogue on one or two tracks. The dialogue editor splits the clips across many tracks so that each has its own processing and its own background. The usual approach, called checkerboarding, places each character's lines, or each camera angle with a different background, on alternating tracks. Clips that need different equalization or noise reduction then sit on separate tracks where the mixer can treat them independently, and the editor can extend each clip's background under the next clip to create smooth transitions.

A common track layout for a feature has six to twelve dialogue tracks, one or two tracks for production effects (sounds from the production recording worth keeping, such as a door or footsteps that sound good), and tracks for room tone fill. The layout should be consistent across reels so the mixer always finds the same material in the same place.

Smoothing backgrounds across cuts

Every shot has a slightly different background because it was recorded at a different time, from a different microphone position or with different noise sources. When two shots are cut together, the background jumps at the edit. The dialogue editor's most important technical skill is hiding those jumps.

The basic technique is to extend each clip's background beyond the edit using the handles and overlap it with a crossfade. A crossfade of several frames, or longer for quiet scenes, lets one background blend into the next. Where the handles contain unwanted sounds, the editor finds clean room tone from elsewhere in the take, or from the recorded room tone, and lays it under the transition. Matching the level and color of the room tone to the dialogue is critical: room tone that is quieter or brighter than the background under the words makes the gaps more obvious, not less.

import numpy as np
rng = np.random.default_rng(310)
fs = 48000
def rms_db(x):
    return 20 * np.log10(np.sqrt(np.mean(x ** 2)))
shot_a = 0.010 * rng.standard_normal(fs)       # background under shot A
shot_b = 0.016 * rng.standard_normal(fs)       # louder background under shot B
print(f'shot A background {rms_db(shot_a):6.1f} dBFS')
print(f'shot B background {rms_db(shot_b):6.1f} dBFS')
print(f'jump at the cut   {rms_db(shot_b) - rms_db(shot_a):6.1f} dB')
fill = 0.012 * rng.standard_normal(fs)          # room tone from take B, a little quieter
a_filled = shot_a + fill                        # uncorrelated noise adds on a power basis
print(f'A with fill added {rms_db(a_filled):6.1f} dBFS')
print(f'remaining jump    {rms_db(shot_b) - rms_db(a_filled):6.1f} dB')

shot A background  -40.0 dBFS
shot B background  -35.9 dBFS
jump at the cut      4.1 dB
A with fill added  -36.2 dBFS
remaining jump       0.3 dB

The code shows the principle behind a common trick: rather than turning shot B down, which would also lower the dialogue, the editor can add a little matching room tone under shot A so that its background rises toward shot B. Combined with a crossfade at the cut, the jump becomes too small to notice. Noise reduction, discussed below, can reduce the louder background, but adding fill is often the more natural solution.

Cleaning: noises, clicks and mouth sounds

Dialogue editors remove a wide range of unwanted sounds: crew noise, footsteps on hard floors during lines, prop clatter, clothing rustle, microphone bumps, lip smacks and clicks, breaths that are distracting, and sibilance. Each removal leaves a gap that must be filled with matching background. Small clicks can be repaired with spectral editing tools that show audio as a frequency-against-time picture, letting the editor paint out a cough or a chair squeak between words without affecting the voice.

Not every noise should be removed. Breaths are part of a performance; taking them all out makes characters sound robotic. Production effects such as a cup put down or a door closed in sync may sound better than anything Foley can create, so the dialogue editor moves them to the production effects track rather than deleting them. The goal is a clean track that still sounds alive.

Noise reduction

Modern noise reduction software, much of it using machine learning, can reduce steady background noise such as air conditioning, traffic hum or camera fan noise, and can separate voice from background with impressive results. Overuse produces artifacts: a watery, swirly sound, loss of consonants, and a voice that sounds detached from its space. Dialogue editors and mixers use noise reduction in modest amounts, often several gentle passes rather than one heavy pass, and compare the result against the original at the same loudness.

Some decisions belong to the dialogue editor and some to the mixer. Many supervising sound editors prefer editors to deliver tracks with only corrective clean-up, leaving broad noise reduction and equalization to the mixer, who can judge it in context with music and effects. Agree the division of labor at the start of the project.

Lip sync and sync checks

Every edit risks moving dialogue relative to picture. Dialogue editors check sync on close-ups by watching consonants that close the lips (b, p, m) and by checking hard sounds such as claps. Common causes of sync errors include clips slipped accidentally while editing, alts dropped in slightly late or early, and frame-rate or sample-rate mismatches during the conform. A one-frame error on a close-up may be visible to attentive viewers; two frames is usually obvious.

Preparing tracks for the mix

A well-prepared dialogue session makes the mix fast. Clips are grouped logically, crossfades are clean, levels are roughly balanced so the mixer is not starting from wildly different gains, unused material is removed or muted, and alternative choices are left on muted tracks nearby so the mixer and director can switch quickly. Notes flag any lines on the ADR list, any production effects that should be kept, and any places where the editor is unsure.

The dialogue editor also prepares the ADR list (lesson 7) as they go, noting lines that cannot be fixed with alts or processing and lines the director wants to change for performance or story reasons. The earlier the list is ready, the earlier actors can be booked.

The ethics of dialogue editing

Dialogue editors can change what people appear to say by moving words from other takes, and in documentary and unscripted work this power must be used carefully. Tightening pauses or removing a repeated word is accepted practice; constructing sentences that a contributor never said, or changing the meaning of a statement, is not. Many broadcasters' editorial guidelines require that editing preserves the meaning and context of what was said, and that audiences are not misled. In fiction, words can be moved freely to serve the performance the director and actor intended, but the actor's work should be respected.

US networks' standards and practices departments and journalism codes, such as the Society of Professional Journalists' code, emphasize accuracy and not misrepresenting what people said.