[wassgha]

Closes #104.

The transcript is rendered as one span per word, so NVDA and other screen readers can't read it as text, report where the caret is, or edit it. This PR adds the "bridge" approach suggested in the issue: a visually hidden native textarea that stays in sync with the transcript. Edits made in it are turned into normal editor actions. The visual layout is unchanged.

How it works

• The textarea comes right after the transcript header in Tab order. It holds the words that will be exported, one line per speaker turn (Alice: Hello there…). Up/Down moves between speakers and word navigation moves between words. Focusing it reads a label and a usage hint.
• Moving the caret seeks the playhead to the word under it. When the textarea gets focus, the caret starts on the word that is currently playing.
• Shift-selecting text selects those words in the editor. The visual transcript highlights them.
• Backspace / Delete cut words. With a selection, every word it touches is cut. With no selection, they cut the word before or after the caret. The cut is announced through a live region (Cut: test).
• Enter opens the existing Correct popup. Focus returns to the textarea when it closes.
• Space plays or pauses. ⌘/Ctrl+Z (⇧ to redo) uses the editor's undo history and is announced.
• Typing is blocked, so the text always matches the edit. Every way of editing (keys, Cut, the Edit menu) is intercepted in beforeinput.
• Sighted keyboard users get an outline around the transcript panel while the textarea has focus.

Changes

• components/TranscriptTextLayer.tsx (new): the textarea, keyboard handling and live-region announcements.
• lib/transcriptText.ts (new): builds the text and maps caret positions to words. Covered by tests/transcript-text-test.ts.
• components/TranscriptPanel.tsx: mounts the layer. Correct now takes word ids and can hand focus back.
• hooks/useTranscriptSelection.ts: ignores the page's text selection while the textarea has focus, so it doesn't clear the words selected from it.
• New strings in all 9 locales. The non-English translations should get a quick review.

Limitations

• Cut words aren't in the textarea, so in text mode they can only be brought back with Undo.
• Split, speaker reassignment and the S / @ shortcuts aren't available from the textarea.

Testing

• npx tsx tests/transcript-text-test.ts, npm run test:i18n, tsc --noEmit, npm run lint
• Drove the textarea with the keyboard in headless Chrome: word and line navigation, selection syncing to word highlights, cutting with and without a selection, blocked typing, undo, Enter → Correct → focus back, Space playing the media.
• Not yet tried with a real screen reader. It would be good to have the issue's author try a build with NVDA.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
app.rescript	Ready	Preview	Oct 5, 2026 12:57am UTC