If room noise rises during pauses in a newly recorded file, check automatic input gain before adding more post-production cleanup. The important distinction is whether the change already exists in the captured source or appears only after an editing effect. Record a short controlled comparison when the recorder provides that setting. Do not assume every volume surge is caused by the same process.
Inspect the captured file first
Keep the new recording and audition it without effects. Choose a passage containing ordinary speech, a quiet ending and a pause. Compare a recording made outside the project if available. If the source already swells, bypassing a timeline compressor cannot restore the original input behavior.
Confirm which microphone is actually selected. An application, operating system or device may participate in the recording chain. Document visible enhancement settings rather than assuming a single switch controls all processing. If the microphone changes between tests, the comparison no longer isolates automatic gain.
Run a single-variable recording comparison
Descript's editor recorder documentation lists Auto gain control as an input-volume adjustment. It also warns about a one-sided input case when a single microphone uses a multichannel interface. Those are Descript-specific details, so changing the control requires checking the recording channels as well as the apparent noise level.
Make two short takes using the same phrase, microphone position and background. Change only the available automatic-gain setting. Use a suitable input level and watch for clipping in both conditions. When no such control exists, record the limitation and investigate the actual capture device instead of inventing a button.
A proposed pause-level worksheet
For a hypothetical narration test, say one normal sentence, leave a natural pause, then say a softer sentence. Save the original files with the setting identified. At comparable speech playback level, note whether the pause background is steady, rises noticeably or changes abruptly. Also record whether quiet words remain complete and whether both intended channels contain the voice.
This is a proposed comparison, not a measured outcome for Descript or a particular microphone. A silent pause alone does not establish success. A test that removes quiet syllables or sends the voice to one ear has traded one fault for another.
Choose a dependable capture setup
Accept a setup when the recording has manageable dynamics, complete speech, no clipping and the intended channel layout. Reject a setting solely praised for an empty noise meter while the voice is missing or unstable. Rehearse a complete short passage before making a long recording with the changed configuration.
Save the approved input device, settings and reference phrase so later sessions can be compared. Existing captured surges may still require a separate external editing decision; turning a recorder setting off now does not undo yesterday's file. Imagild's Skills can help plan the visual material, while capture-level diagnosis belongs to the actual recorder. Keep recording changes distinct from later compression and from any AI-assisted replacement of the spoken content.