Descript: edit a podcast with AI (2026)

Descript treats an episode like a document. Upload the recording, wait for transcription and edit by deleting sentences — audio cuts underneath. In about 25 minutes you leave with an MP3 ready to upload, not an endless timeline project. This flow assumes an already recorded monologue or interview (5–20 minutes); we are not recording the episode from scratch here.

1. Steps 1

   New project and import audio

   Go to descript.com and sign in (web or desktop app). Create a **New project** and pick **Audio project** or **Video project** if you have video too; for voice-only podcast, audio is enough. Use **Import** or drag your file (WAV if you have it; MP3 works). Name it clearly (`episode-12-draft`). Descript starts transcribing once the file is in the project — do not edit until transcription finishes (progress bar or «Transcribing» indicator).

   What you should see: Empty project with your file in the media panel and transcription in progress or already visible under the waveform.

2. Steps 2

   Review the transcript before deleting

   Open the **Script** or **Transcript** view (exact label may vary). Read the first minute while playing audio. Fix proper nouns and technical terms: double-click a misheard word and type the correct form. If ASR split a paragraph oddly, merge sentences only when you are sure — what matters is that **every word you keep sounds right on playback**. Do not cut tangents yet; fix text errors that would change meaning first.

   What you should see: Transcript aligned to the playhead; at least one name or term corrected manually.

3. Steps 3

   Cut tangents by deleting text

   Find a block that will not ship (long intro, digression, repeated «as I was saying…»). Select it in the transcript and delete, or use the **delete / cut** action tied to text (Descript trims the track). Play from a few seconds before the cut to check you are not leaving half words. Repeat for **one or two** blocks in this tutorial — do not try to perfect the whole episode. Habit: content first, micro-edits later.

   What you should see: A clear gap in the transcript and a shorter waveform; playback jumps sound natural.

4. Steps 4

   Remove filler words (then listen)

   In the audio or AI menu, find **Remove filler words**, **Shorten word gaps** or the current equivalent. Descript detects fillers («um», «uh», «you know») per language. Apply once on the voice track and play a section that used to be filler-heavy. If a removal sounds artificial, undo or fix that word manually in the transcript. Do not run the action five times — speech can get unnaturally tight.

   What you should see: Fewer fillers visible in text or marked removed; when listening, rhythm is cleaner without obvious word-by-word chops.

5. Steps 5

   Studio Sound in moderation

   Open **Studio Sound** (sometimes on the track effects panel or under **AI effects**). Enable it on the main voice track. Play a segment with known room noise. If the voice sounds metallic or «exaggerated radio», lower intensity if a control exists or disable and keep only filler removal. Studio Sound reduces noise and improves clarity; it does not replace recording closer to the mic next time.

   What you should see: Studio Sound indicator on the track; less hiss when listening, voice still recognizable.

6. Steps 6

   One bad line: re-record or Overdub

   Find **one** phrase that landed wrong (stumble, wrong fact). Options in Descript:

   1. **Record insert** / re-record at the playhead (if your plan and mic allow). 2. **Overdub** or text correction: select the phrase in the transcript, type the correct version and generate replacement audio with your configured AI voice (needs prior setup and consent; not required on day one).

   For this tutorial, **re-record one short phrase** or leave a «fix in v2» note if you lack Overdub. Play the join between old and new audio — it should sound continuous.

   What you should see: One replaced voice clip or a new take in the same paragraph; no audible click at the splice.

7. Steps 7

   Underlord for a nudge, then export

   If your build shows **Underlord** (AI assistant), use it for one narrow task: «summarize the three topics in this script» or «suggest an episode title» — not to delete content without review. Then **File → Export** or **Publish / Export**: choose **Audio** → MP3 or WAV per your podcast host (many accept MP3 at 128–192 kbps; archive masters as WAV). Name the file with episode and date. The deliverable is audio, not the transcript exported as a blog post.

   What you should see: Export dialog with format chosen and an MP3/WAV in downloads ready to upload.