Wind, hiss, and why both are placement problems

Two problems that sound like microphone faults and are not

Sibilance (the harsh hiss on s, sh, z and t) and plosives (the thud on p and b) are the two artefacts most often blamed on equipment. Both are physical, both are predictable, and both are much easier to prevent than to repair.

A plosive is wind, not sound

Say “peter piper” with your hand in front of your mouth. You feel a puff of air. That blast hits the diaphragm and displaces it far beyond what the sound itself would, producing a large, very low-frequency excursion (often below 80Hz) that can overload the preamp even when the voice is at a moderate level.

Three prevention measures, in order of effectiveness:

1. Speak past the microphone, not into it. Angle it 15 to 30 degrees off the mouth’s axis, or place it slightly above and aimed down at the mouth. The air jet is directional and the sound is not, so you lose almost nothing and avoid the blast entirely. This is free and it is what professionals do.
2. A pop filter, 5 to 10cm from the capsule. The distance matters: hard against the microphone it does very little, because the jet re-forms. Its job is to break up and slow the airflow across a gap.
3. A high-pass filter. The lesson before covers this. It reduces the aftermath rather than preventing it.

Repairing one afterwards: do not apply a global filter for the sake of one word. Automate a steep high-pass (150 to 200Hz, 24dB per octave) over just the 100 to 200 milliseconds of the plosive, or use a clip-gain dip on that syllable. Both are invisible and neither touches the rest of the track.

Sibilance, and what a de-esser is

Sibilant consonants are broadband noise concentrated between roughly 5 and 9kHz: typically 5 to 7kHz for lower voices, 7 to 9kHz for higher ones. Several things make them worse:

• Proximity. Close miking raises them relative to the body of the voice.
• Compression. A compressor turns the loud parts down, and sibilants are not usually the loudest parts, so everything else comes down around them and they stand proud. Compression makes existing sibilance worse, reliably.
• A presence boost. The 2 to 6kHz lift from the EQ lesson lands next door.
• The speaker. Some people are sibilant, and dental work changes it.

A de-esser is a compressor with a filtered sidechain. The detector hears only the sibilant band, so it triggers only on sibilants; the gain reduction is then applied either to the whole signal (wideband) or only to that band (split-band).

• Split-band is transparent for moderate work and is the sensible default: it turns down 6kHz and leaves the voice alone.
• Wideband ducks the entire voice on each “s”, which sounds more natural on heavy reduction because it preserves the relationship between the sibilant and the rest, but it makes the voice dip audibly if pushed.

Setting one without ruining the voice

1. Put the de-esser after compression, not before. Compression is what exaggerates sibilance, so de-essing before it hands the compressor a signal it then re-unbalances.
2. Use the listen or solo-sidechain function to hear what the detector hears. Tune the frequency until you hear the s sounds clearly and the vowels barely.
3. Set the threshold so gain reduction is 3 to 6dB on the worst syllables and zero the rest of the time. If the meter is moving constantly, the threshold is too low and you are dulling the whole voice.
4. Listen for the lisp. Over-de-essing turns “s” into “th”, and it is instantly recognisable once you have heard it.

Two de-essers doing 4dB each, at slightly different frequencies, frequently sound better than one doing 8dB: the same principle as gentle serial compression.

Automation beats processing

For a handful of harsh syllables in a five-minute read, the cleanest fix is not a plugin. Find the syllable, reduce its clip gain by 3 to 5dB, move on. It takes a minute, it is perfectly transparent, and it does nothing at all to the other 99 per cent of the track.

Dialogue editors do a great deal of this by hand, and the reason is simple: a static process applies everywhere, and problems are local.

Breaths, and a word about removing them

Breath noise is the third item people want gone. Resist deleting them entirely: speech without breaths sounds unnervingly inhuman, and listeners cannot say why. Reduce the loud ones by 6 to 10dB and leave them present.

If you do cut one out completely, replace it with room tone rather than silence, for the reason covered in the capture module.

The free path

• Audacity has a De-Clicker and De-Esser in recent versions, and clip-level gain envelopes for manual work.
• The LSP plugin suite includes a proper split-band de-esser, free and cross-platform.
• Calf Deesser is free, LADSPA/LV2.
• ffmpeg can do a crude sidechain-filtered compressor, and adeclip/adeclick repair damage.
• DaVinci Resolve’s free Fairlight page has de-essing built in.
• iZotope RX’s de-click and de-plosive tools are the paid standard; the free options handle a spoken-word track comfortably.

The limitation

A de-esser cannot distinguish a sibilant from any other loud high-frequency content. Cymbals, key jingling, paper rustle and a bright laugh all trigger it, and on a track with music or effects mixed in it will duck them too. This is why de-essing belongs on the dialogue track before the mix, not on the master.

And there is a floor set by the recording. Sibilance that has clipped (very common, because the s is short and can peak far above the vowels) is distortion, not excess level. Turning it down turns down a distorted sound. The only cure is at the microphone, which is why the placement advice at the top of this lesson matters more than any plugin below it.