Edit by text

With a transcript, you can edit speech the way you edit a document. Delete words and that part of the timeline goes; keep only the best lines and the rest goes. Every text edit is an ordinary timeline edit, and each one is a single undo step (Ctrl+Z).

Transcribe a sequence or clips#

1. Open the "Transcript" panel from the left rail. It is also part of the Captions workspace.
2. In "Transcribe with", choose where the speech is recognised:
   • "On this computer": whisper.cpp runs on your computer and nothing is sent anywhere. The model downloads the first time; you're asked before it does.
   • A cloud provider you added, such as Deepgram, ElevenLabs or AssemblyAI, with your own key. To add one, see AI.
3. Choose "Transcribe sequence", or select clips in the timeline and choose "Selected clips".

Transcribing doesn't change your edit. The transcript is kept with the media, so a recording that already has one isn't transcribed again when another sequence uses it.

• Cloud providers ask first. The first time audio would go to a provider, Wabcut asks "Send to <provider>?" and shows what is sent and the price. Choose "Allow once" or "Always allow"; "Cancel" sends nothing.
• Local transcription keeps working offline. Turning cloud AI off, or locking a project to no-cloud AI, doesn't stop "On this computer".
• The graphics card is used when it can be. If it can't, local transcription runs on the processor and tells you.

Find your way around the transcript#

• Click a word to move the playhead there.
• Drag across words to select them. Their range lights up on the timeline.
• Find a word or phrase by typing it in "Find in transcript".
• Speakers. When the transcription tells speakers apart, the transcript is split into paragraphs by speaker.

The transcript shows what you hear in the sequence. A word appears when it sits inside an audio clip, at its place on the timeline, following the clip's speed and direction. Words you trim away leave the transcript, and it keeps up as you edit. Video clips add no words: a linked video and audio pair is heard through its audio, and selecting its words selects both clips.

Delete words#

1. Select the words.
2. Choose "Delete", or press Delete.
3. The words are struck through and the cut shows on the timeline. Choose "Remove" to make the cut, or "Cancel" to leave everything as it was.

The timeline is cut to match, in one undo step:

• Cuts fall on frame boundaries just outside the words, with a little padding on each side (40 ms by default). They never cut into a word you keep.
• Deletions close together, with no kept word between them, become one cut.
• The cut goes on the tracks the words are heard on and on their linked tracks. Sync-locked tracks ripple with them; other tracks stay where they are.
• Words on a locked track can't be deleted.

Keep only the best lines#

1. Select the words you want to keep.
2. Choose "Keep only", then "Remove".

Everything outside the kept words is cut, from the first to the last clip that has words. It's a quick way to rough out a long interview or voice-over.

Move a paragraph#

Drag a paragraph and drop it before another one. Wabcut cuts the clips at those words, moves them before the other paragraph, and closes the gap they left. The move is one undo step.

Fix misheard words#

1. Select the words that were misheard.
2. Choose "Fix…" and type the correct words.

The correction changes the text, not the timeline, and is saved with the project. Captions you already made pick it up when you choose "Update from transcript" in the "Captions" tab.

Remove filler words and silences#

1. In the "Transcript" panel, open "Clean up".
2. Set what to look for:
   • "Fillers": filler words such as "um" and "uh".
   • "Silences over": pauses between words longer than this many seconds. The default is 0.8; 0 turns it off.
   • "Padding around cuts": how much to keep around each cut, in milliseconds. The default is 40.
3. Choose "Find". Wabcut lists the fillers and long pauses it finds.
4. Decide for each one whether it goes or stays.
5. Choose "Preview" to see the cuts, then "Remove". All of them are one undo step.

What counts as a filler:

• English: um, uh, er, ah, hmm. "Like", "you know", "I mean", "sort of", "kind of" and "basically" count only when they sound like fillers: with a pause next to them, a comma after them, or said twice.
• Thai: อืม, เอ่อ, เออ, อ่า. แบบ, แบบว่า, คือ, ก็คือ, ประมาณว่า and อ่ะ count only in the same way.

Edit through multicam and nested clips#

Transcribe the sequence as usual. The words of a multicam clip come from its active audio angle and sit at the right times. Nested sequences work the same way.

Deleting words inside a multicam clip cuts that clip in place: the pieces stay multicam clips, so you can still switch angles afterwards. Each deletion is one undo step. For setting up and switching angles, see Multicam.

Edit Thai speech#

Thai works throughout:

• The transcript shows Thai word by word, without spaces, and its lines break only between words.
• "Clean up" knows Thai fillers as well as English ones (see above). Silences are found in any language.
• Captions made from a Thai transcript break correctly, and there is a "Thai-friendly" caption style.

Make captions from the transcript#

Captions come from the transcript. In the "Captions" tab, choose "Generate captions"; after you edit or fix the transcript, choose "Update from transcript". Styles, translation, SRT and VTT files and burn-in are covered in Captions.

If a source has bleeped words, they read as bleeped in the transcript and the captions too. See Masks and redaction.

Transcript edits and clean-up are also commands you can run from a script. See Scripts and agents.