wlphi/prompter
Smallest and most directly useful candidate. Browser speech mode plus a Vosk path for genuinely offline recognition.
A private comparison page for testing three open-source voice teleprompters before deciding what to bring into Content-MCP.
Open each candidate in a new tab, paste this script, then test voice follow, manual controls, mirror mode, and whether it survives going off-script.
| Criterion | Why it matters |
|---|---|
| Voice follow accuracy | Does it recover after pauses, filler, and misreads? |
| Local-only confidence | Can it require native on-device recognition, or does it just use Web Speech? |
| Integration shape | Can the useful parts be moved into Content-MCP without importing a whole product? |
| Filming controls | Mirror, font size, scroll position, keyboard controls, fullscreen. |
| Maintenance risk | Build health, dependencies, license, and amount of unrelated code. |
Smallest and most directly useful candidate. Browser speech mode plus a Vosk path for genuinely offline recognition.
React/Vite app focused on automatic scroll while reading. Good benchmark for matching robustness and prompter ergonomics.
Full product-style PWA with voice control, camera recording, script history, mirror mode, and mobile-oriented UX.
| Candidate | Best reason to use | Risk to watch |
|---|---|---|
| wlphi/prompter | Closest to a reusable implementation, including an offline Vosk route. | Vosk mode needs model/server decisions if you want it inside Content-MCP. |
| jlecomte | Simple voice-follow teleprompter UX, useful as a matching benchmark. | Source build is currently incomplete in the cloned repo because root index.html is missing; the shipped dist/ works. |
| kosuvorov | Most complete user-facing product and mobile/PWA experience. | Much larger surface area, includes marketing pages and tracking in upstream build; copied analytics scripts were stripped here. |