A cut that goes to digital silence sounds wrong in a way the room’s own hum never does: the ear hears the nothing. This step learns what the room sounds like and puts it back wherever the recording goes quiet.
The room
When a file loads, its quietest second is measured as the room: 96 band levels from 20 Hz to 20 kHz, the same profile the noise reducer learns. Select a stretch of nothing but room and press Use selection as room tone to learn from that instead.
Fill under
A stretch quieter than this for 30 ms is dead air and gets the room. -60 dBFS catches silenced and cut stretches and leaves quiet speech alone; raise it towards -40 dBFS for a recording whose pauses were gated down rather than cut.
Level and fade
Level offsets the fill from the learned room, -12 to +6 dB, for a room that should sit a little under its own recording. Fade is how long the room takes to come up and go away, 5 to 200 ms; 30 ms is unheard on speech.
Example
An interview cut together from two takes, each cut landing on digital silence between words. With the room learned from the quietest second and the threshold at -60 dBFS, each silence is filled with noise whose 96-band profile sits within 2 dB of the learned room from 100 Hz to 10 kHz and within 1 dB overall, measured on forty rooms from read speech; the words either side are untouched, bit for bit.
Where it stops
The room is noise shaped like the room: a hum at 50 Hz comes back as a hum, but a fridge that switches on and off, or a passing car, is not a room and is not reproduced. A room learned from a stretch that held speech fills the gaps with a speech-coloured hiss; learn again from a cleaner stretch.