Every day, speech crosses language boundaries. Conference keynotes dubbed for global audiences. Films re-voiced for new markets. News broadcasts simultaneously interpreted. Presidential addresses translated in real time. The words change. The meaning is supposed to hold. But does it? Not "is it fluent" — does it actually land in the brain the same way? That question has never had a scientific answer. Human reviewers catch errors. Grammar checkers catch structure. Nothing measures whether the cognitive experience of the original survives the crossing. Until you ask the brain directly. That's what langdrift does. Feed it two audio files — original and translation, in any language, any format. It maps both through a model trained on fMRI recordings from 720 people, compares the cortical activation patterns across 20,000 brain vertices, and gives you a drift score. A perfect diplomatic translation: 2.3%. A completely different audio signal: 11.9%. The brain knows the difference even when the content sounds superficially similar. Movie dub feel off but you can't explain why? Measure it. Conference interpretation losing the speaker's authority? Measure it. Localised ad campaign not landing the same way? Measure it.

Langdrift

Detect cognitive drift in speech translations

Detect cognitive drift in speech translations