Descript Studio Sound: reduce noise without cutting off speech

Decide which words must survive the cleanup

A quieter recording can still be a worse recording. Losing the final word of 'that is not allowed' changes the meaning. Pick a passage containing quiet speech and short replies before evaluating the effect.

Our suggested listening sample includes normal speech, a softer voice and words that overlap background noise. Listen once without reading the transcript: knowing the sentence in advance can make missing words easier to overlook. Then try the speaker or headphones your audience is likely to use.

There is no universal best Intensity percentage

Descript recommends lowering Intensity when Studio Sound suppresses speech. Heavy background noise can make separating the voice difficult.

Try a few spaced settings, such as 100, 75 and 50, on the same sentence. These are example comparison points, not tested recommendations. Listen for the opening consonant, the last word and unnatural changes between phrases. Some background sound may be preferable to missing speech.

Make a separate file to treat only one section

Studio Sound applies at source-file level, including other uses of that file in the project. A timeline cut alone does not isolate the effect. Copy the section into a new composition and use Flatten tracks to create a new file before applying it.

Keep an unflattened copy so that later changes to cuts and source clips remain manageable. When flattening, choose Script track only or All tracks according to what you intend to combine. Avoid accidentally baking other media into a voice-only repair.

When you replace the original section, check its duration and insertion point. Play across both joins, not just the repaired sentence. A sudden change in loudness or room sound may be more distracting than the original noise.

Budget for variable AI-credit use

Current documentation caps manual Studio Sound at 30 AI credits per file application; it is not a fixed ten-credit action. Cost varies with the job. Rerunning an AI operation charges again, and Underlord can add its own usage.

Uploaded media minutes and AI credits measure different things. Select the relevant Drive in Settings and check Usage. Decide which files and sections need treatment first, process one, inspect the result and usage, and then proceed to the rest.

Treat echo and synchronization as a separate problem

For certain recordings, including affected screen captures, Descript documents echo or drifting synchronization after Studio Sound. Repair audio drift is available from the affected file's menu in Project, Files.

Compare timing near the beginning, middle and end. Note whether the offset grows or stays constant. Keep a source copy before making repairs, then export a short result to check outside the editor. Repeated Intensity changes are unlikely to diagnose a timing problem.

Finish audio repairs before final caption delivery

If an audio repair requires rebuilding a section, do that before finalizing subtitle files. Cutting speech after delivery can also change subtitle timing.

A practical order is to finish the audio edit, correct captions, then export a matching video and subtitle pair. Our captions guide covers that next step. Audio enhancement does not establish that the source language is supported for automatic transcription; check language support separately.