Fixing uneven dialogue levels during editing

Fix uneven dialogue by identifying why the level changes, adjusting clips or phrases, then measuring the finished mix. Do not raise every quiet passage automatically: some contain low speech, while others are quiet because the microphone missed the voice and captured mostly room noise.

Editing tools may include clip-level balancing, loudness matching across audio files, and compression effects. Use them within the same diagnosis-to-final-check workflow, rather than treating a matched level as proof that dialogue is clear.

Diagnose the variation

Listen across cuts at a steady playback volume. Mark where the same speaker changes level, one speaker consistently sits below another, or a word drops because the person turns away. Check whether any section clips or distorts; lowering a clipped recording will not restore lost detail.

Compare the source tracks. A clean low-level recording can often be raised. A distant recording brings echo and noise up with the voice.

If an alternate mic or take exists, it may be better than heavy processing. Keep the original file and work non-destructively so each adjustment can be reversed.

For clean recordings of the same speaker, compare levels across audio files. Listen to the clips in sequence as well: the reading does not reveal whether a take sounds distant, noisy or distorted.

Use a sample-peak meter in dBFS to check the highest sample level; use LUFS or LKFS to assess perceived loudness. True peak, measured in dBTP, estimates inter-sample peaks that a sample meter can miss. For the same BS.1770 measurement, LUFS and LKFS use the same numerical value.

Key Audio Measurement Units in Dialogue Editing

Balance before compressing

Use clip gain or volume automation to bring whole takes and individual phrases into a sensible relationship. Apply these adjustments in your editor, checking the result across the full passage.

In your editor, use clip-volume controls to balance takes, then adjust automation for phrases that still stand out or fall back. Make changes in context across the passage; do not rely on an unsupported fixed gain amount when a quiet clip may contain noise.

After basic balancing, apply gentle compression if the voice still swings sharply between soft and loud words. Your editor's compression tools provide a place to apply it, but no single setting suits every recording; compare the result with the unprocessed voice.

Compression can also make breaths, room hiss or mouth noise more prominent, so listen for these while adjusting it. A noise gate can cut off quiet syllables if set aggressively, so test a full sentence rather than one isolated word.

Check the whole programme

Dialogue needs to work with music, effects and voiceover. Reduce competing tracks while speech is present instead of pushing dialogue to a harsh level. Measure the final export against the destination's current loudness requirements where there are any. Do not assume one broadcast target applies to every web video.

Loudness matching across multiple audio files can check dialogue clips against one another. For the finished programme, use the destination's stated requirement where one exists. Delivery figures are not dialogue-clip settings.

Published programme examples include EBU R 128 at -23 LUFS, generally no higher than -1 dBTP. ATSC A/85 sits at -24 LKFS for content without matching metadata, with about ±2 dB measurement tolerance and below -2 dBTP in the cited guidance. These are destination examples, not a universal dialogue level.

Apple Podcasts is listed at about -16 dB LKFS, ±1 dB, with true peak no higher than -1 dBFS. Netflix uses its current title-specific delivery specification. YouTube's creator formatting guidance has no public fixed upload LUFS requirement.

DaVinci Resolve's Fairlight page offers selectable standards, tolerance colour coding, a main-bus loudness graph and offline clip analysis. Check the exported programme on the main bus with the applicable loudness measurement, then listen to the export from start to finish on headphones and a phone speaker.

The practical test is whether a viewer can understand every speaker without reaching for the volume control between cuts. If a source line remains unintelligible, use a retake, an alternate track or an editorial rewrite rather than claiming that level matching repaired it.

Loudness Standards for Different Platforms (Australia & International)