How to Edit a Podcast in Descript: Audio Cleanup, Filler Words, and Loudness Workflow

Overview

Descript podcast editing starts with a transcript. After you import or record an episode, let the transcription finish and review the text against the audio before making major edits. Transcript accuracy matters because deleting a sentence, changing a speaker label, or removing a repeated phrase can affect the underlying audio edit.

A reliable sequence is:

1. Organize the project. Name the episode, confirm the sample rate and source files, and duplicate the original before making substantial changes.
2. Make editorial cuts. Remove mistakes, repeated answers, long tangents, and sections that do not serve the episode.
3. Clean speech. Address filler words, excessive pauses, false starts, and obvious interruptions.
4. Repair the recording. Reduce distracting room noise or gaps, and check that voice processing has not introduced artifacts.
5. Balance the mix. Match speakers, set music beneath speech, and smooth transitions with short fades or crossfades.
6. Quality-check and export. Listen away from the editor, review loudness and peaks, and export the format required by your hosting or publishing workflow.

The goal is not to remove every imperfection. Natural breaths, brief pauses, and occasional informal wording can help a conversation feel human. Edit anything that distracts from meaning, rhythm, or intelligibility, then preserve enough variation to avoid a tightly compressed or artificial sound.

Checklist by scenario

For transcript-based cuts

• Read through the transcript once before editing so you understand the episode’s structure.
• Mark the opening, key segments, sponsor or announcement sections, and closing.
• Delete complete thoughts or sentences where possible instead of removing isolated words from the middle of a phrase.
• Play across every cut. If the rhythm jumps, restore a few words, shorten the cut, or add a small transition.
• Check speaker labels before applying edits to a multi-person recording.

Transcript editing is most effective for content decisions. It is less reliable as a substitute for listening: transcription may miss a quiet word, misidentify a speaker, or fail to represent a pause accurately.

For removing filler words

• Use Descript’s filler-word detection or transcript search to identify repeated “um,” “uh,” “like,” and similar expressions.
• Review each suggested removal in context. Keep a filler word when deleting it would create an abrupt rhythm or change the speaker’s meaning.
• Remove clusters of fillers around a restart, but preserve a natural pause when the speaker is thinking or emphasizing a point.
• Listen to the edit with headphones and speakers. Glitches that are subtle on headphones may be obvious on a laptop or phone.

For a natural result, treat filler-word removal as selective cleanup rather than an automatic target. A few verbal habits can be part of a person’s voice; removing all of them may make the interview sound unnaturally edited.

For pauses, gaps, and false starts

• Shorten silence that feels accidental, especially after a technical interruption or a cut in the conversation.
• Keep pauses that help listeners process a complex answer.
• Remove the abandoned beginning of a sentence only when the replacement joins smoothly to the next phrase.
• Use small fades or crossfades at edits that produce clicks, pops, or sudden changes in room tone.

A useful test is to listen without watching the transcript. If you can identify the edit immediately, make it less aggressive or smooth the transition.

For room noise and audio repair

• First identify the source: steady HVAC noise, traffic, electrical hum, keyboard sounds, or inconsistent microphone distance require different treatment.
• Apply noise reduction conservatively. If the voice develops watery, metallic, or underwater artifacts, reduce the intensity or use the cleanest available source recording.
• Use voice enhancement or repair features only after the editorial cuts are stable. Processing a clip that will later be deleted wastes time and can make comparison harder.
• Compare repaired audio with the original at the same perceived volume.

Noise repair cannot fully recreate missing detail. When a section is badly damaged, a clear edit, brief explanation, or replacement recording may sound better than heavy processing.

For two or more speakers

• Check that each person has the correct speaker label and track.
• Match perceived voice level rather than relying only on meter readings.
• Listen for differences in room tone, microphone distance, sibilance, and low-frequency buildup.
• Do not process every speaker identically if their recordings have different problems.
• Review interruptions and overlapping speech manually; transcript-based cuts can make overlaps feel unnatural.

For music, intros, and outros

• Keep speech clearly dominant over music.
• Fade music in and out rather than starting or stopping it abruptly.
• Leave enough space before the host begins speaking so the opening does not feel rushed.
• Check the transition from music to dialogue on small speakers, where low-level background music may compete with consonants.
• Confirm that you have permission to use any music included in the final episode.

What to double-check

Speech consistency

After cleanup, compare the beginning, middle, and end of the episode. Watch for sections that are noticeably louder, brighter, thinner, or more compressed. Speaker consistency matters more than making every moment identical. If one guest is much quieter, adjust that section carefully rather than raising the whole episode until louder sections peak.

Loudness and peaks

Podcast loudness recommendations can vary by platform, format, and publishing workflow, so use the target supplied by your host or distribution process when one is provided. As a general workflow, inspect integrated loudness, true or sample peaks when available, and the loudest transitions. Do not judge loudness from waveform height alone.

Before export, listen to the episode at a comfortable level and check:

• Speech is intelligible without constant volume adjustments.
• No word, laugh, breath, or music hit causes an unpleasant peak.
• Noise does not become obvious during pauses.
• Intro and outro music do not overpower speech.
• The final minute sounds consistent with the rest of the episode.

Export settings

Use descriptive filenames such as show-episode-number-final-review and reserve “final” for a version that has passed the listening check. Keep the original recording and an earlier edit version so you can recover a section if a later change goes too far.

Common mistakes

• Deleting every pause. This removes conversational rhythm and can make speakers sound rushed.
• Accepting automated filler-word edits without listening. Detection is a starting point, not a final editorial decision.
• Overusing noise reduction. A quiet artifact-filled track is often less pleasant than a lightly noisy natural recording.
• Editing only while watching the transcript. The written text cannot reveal every click, breath, level shift, or room-tone change.
• Normalizing before finishing the edit. Major cuts and music changes can alter the balance, so do final loudness checks after the structure is locked.
• Ignoring transitions. Clean words can still sound disconnected when room tone, microphone distance, or timing changes abruptly.
• Exporting once and assuming it is correct. Play the exported file, not only the project preview, and confirm that it opens and sounds as expected.

For a broader episode workflow, see the Descript podcast editing workflow. If the episode is also filmed, the multicam video podcast workflow can help with camera selection and synchronization. You can also use the finished transcript for show notes, blog posts, and social copy.

When to revisit

Run this checklist for every episode, but revisit the workflow whenever the recording conditions change. A new microphone, remote guest, recording room, interview format, or music bed can change the cleanup decisions that worked previously. Create a short reference episode or test segment when you change your setup, then compare the untreated and processed versions before applying the approach to a full recording.

Review Descript’s transcription, audio repair, filler-word, and export controls when the product interface or feature behavior changes. Also review your hosting or distribution requirements before a major publishing cycle, since requested formats and loudness targets may differ between workflows. Recheck your process at least every six months, or sooner if listeners report inconsistent volume, distracting edits, or playback problems.

Before publishing, complete this final pass:

1. Read the episode from start to finish in the transcript.
2. Listen to every edit without looking at the screen.
3. Check speakers, music, pauses, and room noise.
4. Inspect loudness and peaks using the available meters or your host’s requirements.
5. Export, play the exported file, and confirm the filename and version.
6. Save the project and source files before publishing.

That process keeps Descript’s speed advantage while leaving the final decisions with the editor—where context, taste, and a careful listen matter most.