Why there isn't a microphone in this demo
The Whiteboard's voice posture isn't a setting. It's the design. The decisions below are why this page exists where a record button would be on every other product.
What ships in production
- On-device speech-to-text. When a child speaks, the audio is captured in volatile memory, run through STT on the device, and immediately discarded. Only the resulting text reaches the server.
- No raw-audio fallback. If on-device STT is unavailable on a device, the voice feature is disabled — there is no degraded path that ships audio to the cloud.
- Singing exercises use on-device pitch/timing encoding so the tutor can give feedback without storing a voiceprint.
- Local practice channel writes audio only to the device sandbox. Parent unlock is required for any external share.
What ships in this prototype
Text only. The server has no audio-receiving endpoint. The schema has no audio columns. The privacy strip at the bottom of every lesson screen says so plainly to anyone watching — including parents in the room.
Voice-cloning scams targeting families now need as little as a few seconds of clean audio of the target. Generic chatbots that record children produce exactly that audio. The Whiteboard does not. The choice is not a feature; it is the substrate.
Try it — live
This page uses the same on-device SpeechRecognition API the lesson UI uses. Audio never leaves your browser; only the transcript appears below.