{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/ecd5ec249a224d768746ad1f9f3160db\" frameborder=\"0\" width=\"1600\" height=\"1200\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1200,"width":1600,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1200,"thumbnail_width":1600,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/ecd5ec249a224d768746ad1f9f3160db-c9a368bbd323a0e6.gif","duration":132.2,"title":"Voxarah Demo: We Excuse Dialect. We Never Excuse Decoding.","description":"Voxarah Capture, built for the Meta AI Developer Hackathon.\n\nTwo children can read the same words correctly, and one can still be scored wrong because of how they speak. This demo shows the capture layer separating dialect difference from reading error: a standard literal score, the same reading rescored by Voxarah's deterministic rules engine against published dialect research (Green, Wolfram and Schilling, Thomas), and a genuine miscue still caught. We excuse dialect. We never excuse decoding.\n\nPrivacy: transcription runs on device. No child audio or transcript leaves the device. Meta Model API tooling is used on the passage and material side only, in Passage Studio, where every generated passage must pass Voxarah's own verifier before it can be used.\n\nLive build: https://accent-aware-reading-lab.lovable.app\nSource: https://github.com/Voxarah/accent-aware-reading-lab"}