Step one: speech to text
Everything starts with a transcript. Descript transcribes each file after it finishes processing and uploading, and says it uses third-party AI models that it tests and monitors. [3] Only speech is transcribed; Descript states that music and song lyrics are not supported. [3]
Accuracy sets the ceiling for the whole workflow. Descript says accuracy depends on recording quality, accents, background noise, mic placement, overlapping speakers and the model, and that it can reach up to 95% with clear audio. [3] On a 30-minute interview at an assumed 150 words per minute, that is about 4,500 words, so even at the best case you should expect around 225 words to need checking.
Fixing transcript typos is a separate action from editing media. In Descript, Correct changes the text on screen without altering the audio or video underneath, after which the app re-analyses the media to adjust alignment. [8] Add names and jargon to the transcription glossary before you transcribe to reduce the fix-up work. [3]
Step two: word alignment, the part that makes it work
A transcript alone is just text. What turns it into an editing surface is alignment: each word is tied to the moment its sound begins and ends. Descript says it automatically aligns the transcript with the script audio, so deleting a word removes only its matching audio. [4]
You can see that alignment in the wordbar, which shows which words each section of the waveform represents. When a word is misaligned, you fix its start or end boundary by zooming into the timeline and dragging it, one word at a time. [4] Bad alignment is the main reason a text cut clips the first consonant of the next word.
The wordbar is also where pauses live. Dragging a word left shortens the gap before it and may trim media; dragging right lengthens it, and Descript inserts a gap clip for audio or a freeze frame for video. [4]
Step three: how a text deletion maps to the timeline
When you select a sentence and press Delete, the editor looks up the start time of the first word and the end time of the last word and removes that range from the composition. Descript calls this non-destructive: deleted media stays in the project, hidden from the composition, and can be brought back with Restore removed media. [1]
There are three different ways to make text go away, and they do different things. Delete removes text and media from playback, exports and share pages. [2] Ignore strikes the text through, keeps it visible in the script and skips it in playback; ignored text is excluded from subtitles but can appear in exported transcripts. [2] Remove from transcript does the opposite of what people expect: it keeps the audio and video and hides only the words. [1]
Cut and paste moves media too. Cutting a paragraph and pasting it elsewhere moves the matching footage with it, while copy duplicates it. [1] That is how you restructure an interview by dragging answers into a better order.
| Action in Descript | Text | Audio and video | Subtitles |
|---|---|---|---|
| Delete | Removed | Removed from playback and exports; restorable [1][2] | Removed |
| Ignore | Shown with strikethrough | Skipped in playback [2] | Excluded [2] |
| Remove from transcript | Hidden | Kept [1] | Hidden from captions [1] |
| Correct | Changed | Unchanged [8] | Uses corrected text |
| Filler word Ignore | Strikethrough | Audio removed despite the name [6] | Excluded |
Where text editing breaks
Crosstalk. When two people talk at once on one microphone, the transcript can only list one word at a time, so a cut that removes one speaker removes both. Recording each person on a separate track helps: in a Descript sequence, script edits apply to all tracks together so they stay aligned, and the Sequence Editor lets you remove crosstalk from one speaker or fix a single track. [5]
Music and sound without words. Because only speech is transcribed, a music bed, a pause for a demo or a sound effect has no text to select. [3] Those edits happen on the timeline, which Descript describes as where you finesse timing, visuals, audio sync, layering and pacing. [10]
B-roll and overlays. In Descript a visual layer added to a scene attaches to that scene and ends at the scene boundary by default; a layer can be extended across scenes by dragging its edge or by applying changes to every scene where the layer appears. [9] Descript documentation we reviewed does not spell out how every layer reacts when you delete script text under it, so after a heavy text pass, scrub each scene that carries B-roll and check that cutaways still cover what they should.
Automatic cleanup tools are blunt instruments. Remove filler words works on English transcripts only and offers an Avoid harsh cuts option that skips fillers it cannot remove cleanly. [6] Shorten word gaps lets you target gaps longer than a threshold and shorten them to a set length, such as 200 ms in Descript example. [7] Review results before applying them to the whole composition.
Worked example: a 30-minute interview down to 12 minutes
The numbers below are an illustration of how the passes add up, not measurements from a real project. The order matters: do structural cuts first, when you are reading for meaning, and leave automated cleanup to last, when the remaining material is worth polishing.
Pass 1, structure: read the transcript once and delete the warm-up chat, the off-topic tangent and the repeated answer. Each is a paragraph-sized selection and a single Delete. Pass 2, order: cut and paste the strongest answer to the top. Pass 3, tightening: remove false starts sentence by sentence. Pass 4, cleanup: run filler removal with Avoid harsh cuts on, then shorten long gaps. [6][7] Pass 5, timeline: fix the music bed, check B-roll scenes and listen to every cut point at normal speed.
| Pass | Action | Removed (example) | Running length |
|---|---|---|---|
| Start | Transcribed interview | 0:00 | 30:00 |
| 1 Structure | Delete warm-up, tangent and repeated answer | 11:00 | 19:00 |
| 2 Order | Cut and paste strongest answer to the top | 0:00 | 19:00 |
| 3 Tighten | Delete false starts and weak sentences | 4:30 | 14:30 |
| 4 Cleanup | Filler words and gaps longer than about 1 second | 2:00 | 12:30 |
| 5 Timeline | Trim music, fix B-roll and transitions | 0:30 | 12:00 |
Checklist before you trust a text edit
Text editing is fast because you read instead of scrub. These checks catch the problems reading cannot.
- Record each speaker on a separate track when you can, so crosstalk is fixable. [5]
- Add names and jargon to the glossary before transcribing. [3]
- Use Correct for transcript typos; use Delete only when you want the media gone. [1][8]
- Use Ignore while drafting, then decide later; ignored text is skipped but restorable. [2]
- Run filler removal with Avoid harsh cuts on and review the list. [6]
- Listen to every cut at normal speed and fix clipped words in the wordbar. [4]
- Scrub scenes with B-roll, music or on-screen text after big deletions. [9]
Questions before you decide
Does deleting text in Descript delete the original recording?
No. Descript describes editing as non-destructive: deleted media stays in the project but is hidden from the composition, and you can bring it back by right-clicking the edit boundary and choosing Restore removed media.
What is the difference between ignore and delete in Descript?
Delete removes the text and media from playback, exports and share pages. Ignore keeps the text visible with a strikethrough and skips it in playback; ignored text is excluded from subtitles but can be included in exported transcripts.
Can text-based editing handle music and sound effects?
Not from the script. Descript transcribes only speech and states that music and song lyrics are not supported, so music beds and sound effects have no text to select and must be edited on the timeline.
How accurate are the transcripts?
Descript says accuracy can reach up to 95% with clear audio and depends on recording quality, accents, background noise, mic placement and overlapping speakers. On long interviews, plan time to correct names and misheard words.
Sources & verification
Product details and prices can change. Check the linked provider before buying.
- Descript Help: Edit like a doc Accessed 30 Oct 2026
- Descript Help: Deleting vs ignoring script text Accessed 30 Oct 2026
- Descript Help: Automatic transcription Accessed 30 Oct 2026
- Descript Help: The wordbar Accessed 30 Oct 2026
- Descript Help: Multitrack sequences overview Accessed 30 Oct 2026
- Descript Help: Remove filler words Accessed 30 Oct 2026
- Descript Help: Shorten word gaps Accessed 30 Oct 2026
- Descript Help: Correct your transcript Accessed 30 Oct 2026
- Descript Help: Control how layers behave across scenes Accessed 30 Oct 2026
- Descript Help: Timeline overview Accessed 30 Oct 2026
Sources link directly to providers. Product buttons may use separately labeled affiliate links. Read our disclosure.
