Noise Gate Settings That Do Not Chop Your Words

Noise Gate Settings That Do Not Chop Your Words

A badly tuned noise gate is more noticeable than the noise it removes. The tell is a clipped first consonant on every sentence, and once a listener hears it they cannot unhear it.

Start with the room, not the gate

Measure your noise floor first. Record ten seconds of silence and read the level. If the floor sits at -55 dB, a gate closing at -45 dB removes the noise without touching speech. If the floor is at -35 dB, the room is the problem and no gate will fix it.

Thresholds that work

Set close threshold 6 to 10 dB above the measured noise floor, open threshold 4 dB above close, hold 150 ms and release 200 ms. The hold value is what prevents chopping between words.

Parameter Typical value Effect
Close threshold floor + 6 dB Silences the room between sentences
Open threshold close + 4 dB Prevents rapid on/off chatter
Hold 150 ms Bridges short gaps inside words
Release 200 ms Natural tail instead of an abrupt cut

When not to gate at all

For music, ASMR or anything with intentional quiet passages, a gate destroys the performance. Use downward expansion or broadband noise suppression instead. The WO Mic guide covers the alternative chain for those cases.

Verifying the result

Record thirty seconds of normal speech with pauses and listen on headphones. If you hear the room come up between words, the gate is not closing far enough; if you hear consonants disappear, the open threshold is too high.

Frequently Asked Questions

Gate before or after noise suppression?

After. Suppression lowers the floor, which lets the gate use a lower threshold and sound more natural.

Why does my gate open on keyboard clicks?

Clicks are broadband transients above the threshold. Move the microphone or use a high-pass filter before the gate.

Does the gate affect latency?

Not measurably. Look-ahead gates add a few milliseconds; standard gates do not.

Leave a comment

Your email address will not be published. Required fields are marked *