Fix Muddy Dialogue (Part 2): Compression & De-Ess Without Ruining Clarity

Main ideas / lessons

• Goal of spoken-word mixing (after EQ & noise reduction): Use compression and de-essing to manage dynamics—improving clarity and polish while preserving intelligibility and avoiding audible artifacts.
• Risk of over-processing: Compression and de-essing can make dialogue less clear by:
  • Smashing transients (attack too fast) → consonants lose definition.
  • Excessive compression (threshold too low / ratio too high) → pumping/ducking, reduced headroom over other layers, and less “oomph” to cut through.
  • Over-deessing → lisping or “Daffy Duck” / lispy artifacts.
• Processing should be as invisible as possible: Like EQ/noise reduction, the aim is improved sound without you really hearing the processor working. You want control of peaks/sibilance, not obvious effects.

Method / workflow (detailed steps)

1) Choose where compression lives in the chain

• Prefer putting compression after:
  • EQ
  • Noise reduction (if needed)
  • De-essing (DS)
• Compression and de-essing interact, so order matters.

2) Track compression vs bus compression

• Track compression (often bypassed in this approach):
  • Used differently than bus compression.
  • Can be more aggressive/bitey (faster attack) if enabled.
• Bus compression (main recommended approach):
  • Typically very gentle, controlling peaks while keeping dialogue transparent.
  • Helps achieve professional loudness consistency without constant manual fader rides.

3) Use bus compression as a “peak catcher”

• Mindset/example starting values:
  • Threshold: around -6 dB
  • If dialogue averages around -27 dB, the compressor mainly affects super loud peaks (≈ 10 dB above average).
  • Ratio: low (gentle)
  • Makeup gain: tiny amount (as needed)
  • Knee: set so compression engages slightly earlier if desired
    • A bigger knee = smoother start, but can increase reduction.

Attack and release (important for intelligibility)

• For dialogue, attack and release are crucial.
• If attack is too fast:
  • It clamps down consonant transients (e.g., B/P starts; consonants lose “bite”)
  • Result: lowered clarity; consonants sound “smooshed” and less defined
• Preferred guidance (examples mentioned):
  • Slower attack: ~100 ms
  • Slower release: ~200 ms
• Language note:
  • Some languages rely more on transient clarity (examples: German, Spanish, and discussed broadly Mandarin).

4) (Optional) Track compression approach (if used)

• When using track compression, the described strategy:
  • Faster attack: ~5–10 ms (more aggressive/bitey)
  • Threshold: around average levels (example: -24 dB) or left higher
  • Ratio: may be increased vs bus
  • If available, use range/max reduction limiting:
    • Keep reduction small so shouts don’t become crunchy
    • Example intent: cap compression around ~6 dB max

5) Listen critically; avoid audible pumping/gain reduction

• The speaker contrasts “bad compression” vs “good compression” by changing threshold/ratio/attack and comparing.
• Target:
  • When bypassed, dialogue should already feel contained
  • You should not hear pumping or obvious gain reduction
  • Compression should be transparent—you should hear the story, not the effect

6) Know when compression hurts clarity (esp. with music/SFX)

• Mechanism described:
  • Heavy compression reduces dialogue peak headroom.
  • Music/SFX fill up the low end, so dialogue peaks can’t rise above other layers.
• Guidance for action-heavy sequences:
  • Consider bypassing bus compression, or
  • Raise the threshold / lower the ratio so dialogue retains dynamic range and cuts through.

---

De-essing (DS) methodology

When to use a de-esser

• Use a DSER when every “S” is consistently sharp and irritating.
• If it’s only an occasional stray “S,” use volume automation / fader ducking instead.
• Notes:
  • DS is typically done track-by-track because sibilance can be mic-dependent.
  • Mentioned mic examples: certain lavalier mics (e.g., Sennheiser / similar models), contrasted with DPAs described as smoother.

How to set up DSER (track-by-track)

• Example using FabFilter Pro-DS:
  1. Solo the track/clip with sibilance.
  2. Loop only the problematic “S” moment (don’t process normal speech “air”).
  3. Use the DSER’s band-pass / frequency range to target where the S lives.
     • Adjust for voice gender differences (female sibilance often higher frequency, generally).
  4. Set DSER threshold so it triggers only on harsh S content (avoid triggering on air/normal consonants).
  5. Adjust range/depth:
     • Enough to remove irritation, but not so much that it causes lisping.
     • Too much reduction creates “lisp/Daffy Duck”-like artifacts.
  6. Verify in the composite mix (unsolo context):
     • The S should be less harsh while intelligibility stays natural.
  7. Keep it clip-specific when possible:
     • Bypass except for the sections/moments needing DS.

Demonstrated “wrong way” to DS

• Making the DSER too broad/aggressive:
  • Lower threshold so it triggers on too many sounds
  • Or set maximum reduction too high
• Result:
  • The S gets “smacked down” excessively → lispy artifacts.

---

Practical target numbers / starting points mentioned

• Compression target: aim for ~3–6 dB reduction on the biggest loud peaks (not constant reduction) for transparency.
• Example bus behavior:
  • Threshold around -6 dB
  • Dialogue averages ~-27 dB so compression mainly hits louder peaks
• Track compression (if used):
  • Attack 5–10 ms
  • Cap reduction via range (example intent: up to ~6 dB max)
• Bus attack/release for smoothing/leveling (example):
  • Attack ~100 ms
  • Release ~200 ms
• DSER depth:
  • A specific case may require around ~15 dB reduction, but going that far risks making sibilance unnatural.

---

Closing principles

• Do the least processing necessary to get the biggest result.
• Train your ears:
  • Compression is less obviously identifiable than EQ; listen for cues like “squished/crushed/pumping.”
• Avoid the “fader monkey” mindset:
  • Don’t apply changes automatically just because you can—apply them based on what the source material needs.
• End objective:
  • Support the story with controlled dynamics:
    • not so dynamic that listeners adjust volume,
    • not so crushed that it causes fatigue or poor intelligibility,
    • not so de-essed that it turns lispy.

---

Speakers / sources featured

• Speaker (primary): Unnamed instructor/mixer (referenced only by role; “Chris Nolan” mentioned as a joke/example, not as a featured speaker).
• Software / plugins mentioned:
  • Avid Pro Tools (Pro Compressor, track/bus workflow, ProTools automation approach)
  • FabFilter Pro-C 2
  • FabFilter Pro-DS
  • MCDSP SA2 (de-esser mentioned as an alternative; details briefly described)
  • iZotope (de-essing alternatives mentioned generally)