How to Edit a Podcast by Its Words (Remove Ums and Pauses)

The short version

Transcribe the recording on your own computer, so nothing is uploaded.
Delete ums, uhs and repeated words, then shorten long pauses rather than deleting every one.
Listen to every cut: a cut in the middle of a breath or a word can sound wrong.
Export captions from the same transcript for video.

Transcribe

A speech model writes out what is said, with the time of each word. The transcript is your map. Accuracy depends on the audio: a clean recording gives a near-perfect transcript, a noisy one needs checking. Clean the audio first, then transcribe, for the best result.

Remove the ums and uhs

Ums and uhs are easy to find in a transcript. Cut them, then listen. Some are part of a natural speech rhythm; removing every one can make a voice sound robotic. Keep the ones that sit between thoughts.

Shorten the pauses, do not delete them

Long pauses make an episode drag. Shorten the ones over about 0.8 seconds to a third of a second, and leave the pauses that give a line weight. A pause that falls after a joke or before a change of topic is usually worth keeping.

Check each cut

Every cut is a join. Listen to it at normal volume and listen for a click, a change of room tone or a clipped word. A short crossfade at each cut hides most joins.

Export captions

The same transcript can become SRT or VTT caption files for video platforms, or a plain text file for show notes. Check the captions against the audio before you publish.

What it cannot do

A transcript is only as good as the audio, and speech models can mishear names and technical words. Proofread the names and check any quote before publishing.