Reduce sibilance

Sibilance is the high-frequency sound in consonants such as “s” and “sh.” Those sounds are part of speech. The aim is to reduce an excessive edge while keeping the consonant recognizable.

Listen to a complete sentence, including the pauses. Does the sharp sound follow particular consonants, or does a similar noise continue when nobody speaks? This helps you choose what to investigate; it does not diagnose every recording by ear.

Apple’s DeEsser 2 guidance identifies the voice, microphone interaction and compression settings as possible contributors. A sharp recording does not, by itself, show which one is responsible.

A de-esser is a processor that reduces sibilant sound when its detector responds to the recording. Frequency controls help select what it listens for. The threshold determines when reduction begins, while a range or maximum-reduction control, where available, limits how much it reduces.

The processing mode matters. A broadband de-esser can lower the whole signal during a detected consonant. A split-band or frequency-selective mode targets part of the frequency range. Control names and behavior differ between tools: check the manual for the one you are using.

For example, the iZotope RX 6 manual distinguishes Classic processing, which reduces all frequencies when triggered, from Spectral processing, which targets high-frequency sibilance. That distinction is more useful than assuming every de-esser leaves the rest of a word untouched.

An ordinary EQ cut changes the selected frequency balance throughout the treated passage. Selective de-essing can respond when the troublesome consonant occurs. Either can make speech dull if applied too strongly. Our guide to muffled audio explains the broader tonal adjustment problem.

Dynamic EQ is another option: a selected frequency band can reduce in response to the signal instead of staying cut throughout the passage. Sound On Sound’s explanation compares this with conventional de-essing. It gives more control over the frequency area being reduced, but still needs the same listening checks for damaged speech.

Keep the original unchanged. Make a working copy, retain any earlier processed versions and note the timestamp of the passage you want to improve. Include a nearby phrase from the same speaker that already sounds clear.

1. Listen before changing anything. Choose a sentence containing the sharp consonant and enough surrounding speech to hear how it fits. Listen at a comfortable level. Repeatedly raising playback volume makes a harsh passage harder to assess.
2. Find the relevant frequencies. Use the processor’s listen or audition mode if it has one. Adjust its frequency selection until the troublesome consonants stand out. An analyzer can help locate them, but the visible peak is a guide to listening rather than a setting to copy blindly.
3. Begin with light reduction. Adjust the threshold so the effect responds to the sharp sounds. Limit the reduction where the tool permits it. Change one control at a time so you can hear what changed.
4. Return to normal playback. Turn off detector-only listening. Compare the treated sentence with the original at a similar comfortable volume. The question is whether the speech is easier to follow, with recognizable consonants and a natural voice.
5. Check the clearer phrase. Listen for missing word endings, an unnatural lisp, dullness or a dip in the whole voice. Reduce or undo the treatment if it damages speech that was already useful.
6. Save and check the output. Keep the trial separately named, record its settings and listen to the saved file. Check additional speakers and passages before applying the same treatment throughout. For video, check that speech remains in sync.

There is no single frequency or threshold that fits every voice and recording. A preset is a starting point to preview. The useful setting is the one that improves this passage without sacrificing the detail you need elsewhere.

Back off when “s” begins to sound like an unnatural “th,” consonants seem to disappear, or the voice becomes duller between the sharp sounds. The Softube Deesser manual specifically warns that excessive de-essing can create a lisping sound. If a broadband mode makes the whole word dip, compare a more selective mode where available or reduce the amount of processing.

Listen for pumping: a changing high-frequency level that draws attention to the processing. The RX manual notes that processing speed can either smooth useful transients too much or leave attenuation running too long. Timing controls are therefore another setting to compare, not a reason to keep lowering the threshold.

If only a few consonants cause trouble, try restricting treatment to those regions rather than processing every sentence. Apple documents selection-based processing as an available approach in Logic Pro. Keep enough surrounding audio to check the joins and avoid a sudden change in tone.

The quietest “s” is not automatically the best result. Stop when further reduction makes the recording smoother but the words less distinct. An unresolved word should remain unresolved; a more comfortable sound does not establish what was said.

For a recording you can repeat, make a short test with the same speaker, microphone and room. Compare small changes in microphone angle and distance before recording the full conversation. Antares’ recording guidance recommends testing an off-axis position rather than aiming directly in front of the mouth. Listen to the voice and room sound in each test before choosing a position.

A pop filter is not a universal solution to sharp consonants, and a high-pass filter is not a general de-essing control. Avoid buying equipment or applying a fixed filter just because the recording sounds hissy. First establish whether the problem follows consonants, remains in pauses or appeared during processing.

For an event already recorded, a new microphone cannot change the captured source. Work from the earliest, least-processed available file and preserve it.

Ask for a review when de-essing keeps damaging the voice, you cannot separate the sharpness from other noise, or the passage that matters remains difficult to follow. You do not need to diagnose the fault before asking.

Tell Media Medic what the recording contains, the timestamps that matter, which version is closest to the original and what processing has already been tried. Describe the listening difficulty without suggesting the words someone should hear. Discuss any file-sharing requirements before sending identifiable material.

Media Medic offers a free file review. An audio engineer listens to assess whether restoration can improve the recording before an invoice is sent for the work. Restoration is paid work; discuss the proposed work and quote before deciding whether to proceed.

Request a free file review.