Where the latent manifold meets the user.
The lab's quantitative methods become things people can hold and speak to. Two systems run live in your browser, rendered frame by frame from code: an immersive language tutor that moves through 3D scenes, and a multilingual interpreter that turns speech into speech across languages.
rendered live · no recordings · no servers
VR + AI language learning
A learner walks through three rendered worlds with an AI tutor at their side. Café, airport, meeting room. Each scene sets a task, surfaces the vocabulary, and gives feedback in real time.
Head-tracked scenes place the learner inside the language, not in front of it.
Voices come from where the speaker stands, so listening trains direction and distance too.
The tutor parses each utterance, scores pronunciation, and adapts the next prompt.
Multilingual live interpreter
One device, many tongues. Speech enters on the left, runs through the recognition and translation pipeline, and leaves as a new voice on the right. Watch it relay across English, Chinese, and Japanese.
segment speech from silence
ASR · acoustic + language model
encoder · decoder · attention
TTS · neural vocoder
Rendered in the open
Every frame on this page is computed live in the browser. Nothing here is a recording.