Podcast editing software or a service?

Claude DIY walkthrough

1. 01

   Build a clean source project

   Import one track for each speaker, align the tracks, and label the speakers. Correct names, products, technical terms, and obvious transcription errors before asking Claude to judge the story. A mistaken name is not merely a typo here; it can make the wrong passage look disposable.

2. 02

   Export the transcript Claude can actually use

   In Descript, export a TXT or DOCX transcript with speaker labels and timecodes at paragraph breaks. Descript’s transcript export settings let you choose the timecode interval and include speaker names. Keep the raw audio open beside you; the transcript is a map, not the recording.

3. 03

   Give Claude an editorial brief before the transcript

   State the listener, the episode’s promise, the target runtime, and the one idea the listener should remember. Name any sponsor language, disclosures, quotations, or vulnerable passages that must stay. Tell Claude what may be tightened and what requires a factual or editorial check.

4. 04

   Ask for a paper cut, not a rewritten episode

   Require one row per proposed decision: start time, end time, keep/cut/move/tighten, reason, factual risk, confidence, and whether someone must hear the audio. First ask for the episode’s current argument and proposed structure. Only then ask for line-level cuts. This keeps local trimming from destroying the larger story.

5. 05

   Review every uncertain decision

   Read the cut list against the transcript. Hear the source audio wherever Claude marks low confidence, a speaker interrupts, a joke depends on timing, or a cut changes the apparent meaning. Check names, claims, sponsor copy, music rights, and sensitive remarks yourself. Approve the plan before touching the media.

6. 06

   Apply the approved cut in the editor

   In Descript, deleting or moving transcript text changes the linked audio. The edits are non-destructive and restorable, but work in small groups of approved changes and play across every new boundary. Do not bulk-accept cuts merely because the rows look orderly.

7. 07

   Repair and level after the story is settled

   Remove noise, repair clicks, shorten gaps, and control loudness after the editorial cut. Auphonic can automate leveling, noise and reverb reduction, and encoding; it cannot decide whether a hesitation belongs in the story. Keep an untouched copy of every source track.

8. 08

   Listen to the export from beginning to end

   Check the opening promise, transitions, breath, overlap, room tone, music, sponsor language, and ending. Confirm the file format and loudness required by your host. Apple’s published target is about −16 dB LKFS, ±1 dB, with true peak no higher than −1 dB FS. Your host may specify something different.