Conversation, rendered in
every language —
including silence.
SignSpeak observes your hands with on-device tracking and translates ASL into English subtitles, while transcribing the voice across from you. Both of you read the same screen, in real time.
Latency
≈ 1.8s
per signed phrase
Tracking
21 pts
MediaPipe on-device
Model
Gemini
vision interpreter
Privacy
Local
frames discarded
§ 01 — The Craft
Three instruments, one stage.
Every gesture, every syllable — captured by a different sensor, synchronised on the same line.
Sign → Subtitle
Live hand & finger tracking feeds short, contextual clips to a vision interpreter that captions ASL in English.
Voice → Subtitle
The other person simply speaks. Speech is transcribed instantly, set side-by-side with your signs.
One shared score
A single transcript holds both voices. Tap any line to hear the latest sign read aloud.
§ 02 — Manifest
Accessibility, treated as luxury.
The hearing world has had instant captioning for years. We believe the same fluency belongs to anyone who signs — rendered with the same precision, the same restraint, the same care for the room you're standing in.
SignSpeak is not a translator. It is a companion to one — quiet, meticulous, and built to disappear into the conversation.
§ 03 — A Note on Accuracy
Best-effort, never certified.
ASL recognition uses a general vision model, not a model trained exclusively on sign language. Fingerspelling, short common signs, and clear, deliberate gestures perform best. Treat it as an aid to the conversation — not a replacement for a human interpreter.
§ 04 — Colophon
Crafted by xx4x.
SignSpeak is a study by xx4x AI Labs — an independent studio building privacy-first AI tools, games and chat apps.
Every product processes only the minimum data needed and forgets you the moment you're done. Some experiences require accounts — those get scrambled on deletion.