Command · Remove Fillers and Silences
transcript.cleanup
Removes every filler word and pause the review finds with at least minConfidence, in one undo step.
- v1
- Capability EditSequence
- Tool
transcript_cleanup
Request
{
"command": "transcript.cleanup",
"params": {
"fillers": true,
"silences": 0.8
}
}Parameters
fillersboolean- Look for filler words (um, uh; Thai อืม, เอ่อ ...). Default true.
minConfidencenumber (0.0..1.0)- Lowest confidence removed (0..1). Default 0.5.
paddingnumber (0.0..2.0)- Seconds kept around the cut words (never into the next word). Default 0.04.
roomTonenumber (0.0..5.0)- Seconds of each removed pause kept on both sides. Default 0.15.
sequenceinteger (1..)- Sequence id; only needed when the project has several.
silenceDbnumber (-200.0..0.0)- A pause louder than this (dBFS) has sound in it. Default -40.
silencesnumber (0.0..3600.0)- Pauses at least this many seconds long; 0 = none. Default 0.8.
Discussion
Removes every filler word and pause the review finds with at least minConfidence, in one undo step.
Run it with wabcut-cli cmd <project> --json, from an MCP client as transcript_cleanup, or from an extension. See Edit commands for requests, references and results.
See also
transcript.*
- transcript.addSpeaker
New Speaker. Adds a speaker to the project.
- transcript.assignSpeaker
Assign Speaker. Says who speaks over a timeline range: the words the audio clips there play (on "track" only when given) belong to "speaker", or to a new speaker called "name".
- transcript.correctRange
Correct Words. Fixes several consecutive words at once (they read "text", split over them by spaces).
- transcript.correctWord
Correct Word. Fixes a transcription mistake: the word reads "text" from now on, in the transcript the project sees and in captions generated or updated afterwards.
- transcript.corrections
Transcript Corrections. The project's transcript corrections and how they apply to the current transcripts: applied, conflict (the words changed since, e.g. after re-transcription; the correction still wins), restored (no words there any more; the correction brings them back) or noTranscript.
- transcript.deleteRanges
Delete Words. Deletes words (and pauses) like deleting text: ripple deletes on the tracks the words are heard on and their linked partners, cut on frame boundaries just outside the words, neighbouring deletions merged, sync-locked tracks following, locked tracks untouched.
- transcript.keepRanges
Keep Only Words. Rough cut: keeps the given word ranges and ripple-deletes everything else on the transcript's tracks, from the first to the last clip that has words.
- transcript.mergeSpeakers
Merge Speakers. Makes speakers one: every word, assignment and caption of "from" (one id or several) becomes "into", and the merged speakers go.
- transcript.moveRange
Move Words. Moves words (a sentence, a paragraph) to the word boundary before another word: cut out with a ripple and inserted there, with their linked video.
- transcript.renameSpeaker
Rename Speaker. Renames a speaker and/or changes its colour.
- transcript.review
Review Fillers and Silences. Filler words (language-aware: um, uh, like in disfluent context; Thai อืม, เอ่อ, แบบว่า ...) and pauses longer than a threshold, as a review list: kind, word indices, time range, confidence, label and context.
- transcript.speakers
Speakers. The project's speakers (id, name, colour, words in this sequence) and the diarization labels of the sequence's transcripts with the speaker each one is (null: none yet; transcript.renameSpeaker by media and label names one).
- transcript.text
Transcript Text. What the sequence says, in timeline order, as paragraphs (split at speaker changes and pauses) with the word indices the other transcript commands take.
- transcript.transcribe
Transcribe Media. Transcribes the audio of media items or clips (default: every audio clip of the sequence) so the other transcript commands have words. provider: "auto" (on this computer when the local engine is built, else the user's cloud transcription model), "local", or "cloud:"; cloud transcription goes through the user's consent, spend cap and offline / no-cloud locks.
- transcript.words
Transcript Words. The sequence's words in timeline order as arrays in wordFields order: index, text, timeline start and end, the audio clip and track it is heard on, speaker and recognition confidence.