Cut filler words and long pauses from a spoken take using word timestamps and FFmpeg

Guide ID: art_01M6SNVGQ86GDSESP316R61M6Q

Editorial publication checks are separate from community and evidence review. The evidence labels retain this revision’s reported inspection, execution and review status.

Guide title

Cut filler words and long pauses from a spoken take using word timestamps and FFmpeg

Coverage

For anyone editing a recording of one voice who wants fillers such as um and uh removed and long pauses shortened, including fillers spoken at full level that silence tools keep. You end up with an edited WAV and a numeric check at each join. The guide is documentary: its scripts were not run for this revision, and its figures are author-reported from one synthetic fixture.

Example

Author-reported and unverified: a 12.2 s synthetic tone fixture with a hand-written 29-word transcript, 7 words removed and one pause shortened, giving a 6.325 s result. No speech or ASR output was used.

Useful starting points

• Read the transcript against the take before editing: local Whisper can normalise fillers, so a filler you hear may be missing from the word list.
• A silence tool keeps a filler spoken at full level because it is not quiet; word-level timestamps are what make removing it possible.
• Listen to every join. The numeric checks only catch large bursts, so clicks, clipped words and leftover filler tails need a judgment by ear.