1. Listen
Press Space (or K) to play and pause, J/L to skip 5 seconds. Click any word to play from it; in Settings, Clicking a word can play just the word, or the word with a few seconds around it. Drag across the waveform to scrub (or focus it and use the arrow keys; Home and End jump to the start and end), and slow down with Speed.
To hear a hard passage again and again, press Repeat on its paragraph. For a custom range, open Repeat in the player, use Mark start and Mark end while playing, and tick Repeat on.
2. Correct
Press Edit on a paragraph and type. Typing is kept locally; leaving the paragraph saves an undo step. ⌘/Ctrl + Enter also saves. If browser storage fails, keep this tab open and use Save project. Undo and Redo in the player act on what the open tab edits (the transcript or the captions); History goes back to any earlier version of a paragraph.
Select words in the editor and press Italic, Bold or Underline to style captions (for titles, emphasis, or sounds like [applause]).
3. Know where to look
Words that any engine was unsure about are highlighted; hover over one to see each engine's score and which was unsure. Settings in the player sets how unsure. It is a hint, not a verdict: confident mistakes happen too, so listen to the whole thing.
Suggestions shows where the two transcripts disagree in the selected paragraph. Listen, then Use this if the other version is right. A suggestion to remove words usually means one transcript invented something during a pause or applause.
Once you edit a paragraph its highlights go away (the scores were for the old words), and changed words get estimated timing, underlined in dotted blue.
4. Fix a mistake everywhere
In Find and replace, type the wrong spelling and the right one, then Find. Matches are highlighted in the transcript. Exact matches start selected; similar spellings start unselected because they can be real words (“since” for “sience”). Listen, select the right ones, and Replace selected. One Undo reverts the whole batch.
Add to dictionary remembers the spelling. Dictionary matches then appear as suggestions, and Find all lists every one.
5. Export captions
Use SRT for YouTube for YouTube uploads: YouTube reads SRT as plain text, so styling is removed. WebVTT keeps bold, italic and underline on YouTube. Captions follow the word timings, two lines of up to 42 characters by default. The Captions panel warns about captions with estimated timing or too much text to read in time; check those before publishing.
6. Edit captions one by one
In the Captions tab, start from the transcript’s captions or import an SRT or WebVTT file. Edit the text in place. Change a time by typing it, with the arrow keys (0.1 s; with Shift, 1 s), or press = to use the playhead. Split breaks a caption at the text cursor, timed at the playhead when it is inside the caption. Shift times fixes captions that are all early or late. While playing, the current caption shows under the waveform: check that it appears with the speech. Next to check visits captions with too much text to read in time, long lines, or very short or long display times.
Once you edit captions, Export captions uses that track. Later transcript changes flag generated captions as stale: choose Regenerate from transcript to replace the track explicitly, or Discard edited captions to export directly from the current transcript. Imported captions remain independent.
7. Save your work
Your work is saved in this browser automatically. To move it to another computer or keep a backup, use Save project and later Open project. New project downloads include the source evidence, dictionary, review decisions and history. Choose the matching audio separately; audio is never included in the JSON. Reopen a download once to confirm your backup. Older project files need their matching episode page.