


Applied AI
OpenEar
AGPL-3.0 · public repoLive captions from a soundboard feed to every phone on the local Wi-Fi, served by one ordinary PC: no cloud, no app. Translation, opt-in, runs on a model that covers about 200 languages. Its speech and translation models were both chosen on measurement.
- of benchmark audio behind the speech model
- 124.8 min
- blind LLM-judge judgments, with hidden controls
- 960
- languages the translation model covers, opt-in
- ~200