Cadenza
read-alongThe pure caption engine behind Lattice's read-aloud: prose in, a timed track out — each word carrying a display form and a spoken form, lit as it's said. It owns no audio and no DOM; you inject the clock. Every beat below runs the real engine — silent by default, and with a connected voice each sentence's measured onset re-anchors the highlight to the real audio.
The storyboard
▸ Cloud voice — paste an OpenRouter key to hear Kokoro (and enable the voice picker)
Your key is stored only in this browser (localStorage), same as the Playground — it never leaves your device. Get one at openrouter.ai/keys. Without a key this page uses your device’s built-in browser voice.
▸ Your narration — paste anything and it re-times live
▸ Generated WebVTT — captions + karaoke word timings, the real export
Driven by the real engine (@/lib/cadenza): buildTrack → a timed track, makeReader → the clock→word driver, toVtt → this export. A plain DOM page, no framework — the proof the core is standalone. With real TTS, each sentence's measured onset re-anchors its words so the highlight tracks the actual voice.