TalkSync

Type some text, hit play, watch the words light up as they're spoken.

The text is run through OpenAI's tts-1-hd for speech and whisper-1 for word-level timestamps. Both results are cached in Vercel Blob keyed on a hash of the input, so the same text never gets generated twice.

Stack

• Next.js 14 (App Router) + Tailwind
• Vercel AI SDK v6 (@ai-sdk/openai for speech + transcription)
• Vercel Blob for storing generated MP3s and transcription JSON
• SWR for client-side fetching with debounce + dedupe

Run it

npm i
vercel link        # one-time, links to the Vercel project
vercel env pull    # pulls OPENAI_API_KEY and BLOB_READ_WRITE_TOKEN
npm run dev

Open http://localhost:3000.

Required env vars

Name	Source
OPENAI_API_KEY	https://platform.openai.com/api-keys
BLOB_READ_WRITE_TOKEN	Auto-set by vercel env pull if Blob is provisioned

Scripts

• npm run dev — dev server
• npm run build — production build
• npm run typecheck — strict TypeScript check
• npm run lint — ESLint