SIGN LANGUAGE TO SPEECH
Your signing. In words.
Sign a short Auslan utterance, then review the English translation.
Interface previewSample phrases will be shown. A trained translation model is not loaded.
- 01Set up your camera
- 02Sign, then finish
- 03Review and listen
Your signing space
A little room to sign.
Keep your face, hands and upper body in view.
Your camera starts when you choose.
Keep your face, hands and upper body in view
English translation
A place for your words.
Your translation will appear here
after you finish signing.
Review the wording before sharing.
Off by default. Review each result first: the same kind of confidence check, tested on another research model, answered people who were not signing about as often as signers.
Experimental translations may be wrong. Review each result. This system does not replace an Auslan interpreter.
Conversation 0
Finished translations stay here for this session. They aren’t used to change the next translation.
How it works
- Make room. Enable the camera and keep your upper body and both hands in the frame.
- Sign one thought. Choose a model, select Start signing, then Finish & translate when you’re done. Recording ends at 10 seconds. On a laptop, the translation arrives about 4–5 seconds after Finish for every second signed; the six-model ensemble adds about a second for image features.
- Review the result. Read the English text, play it aloud or copy it. If asked, try signing again.
More than one person in view? The translator follows each person by where their shoulders are, calls them Person A, Person B and so on, and translates the one moving most like a signer. That choice is a guess, so the result says who was translated and, when someone else was in view for most of the utterance, offers a switch. Nobody is recognised by their face or appearance.
Captured frames are sent to the server during recording and processing. Images are not saved. Capture buffers are released after completion, cancellation or disconnection; an in-flight computation may finish first. Turning the camera off stops its live feed. Use Clear to discard the current utterance.
Session details
- Models loaded
- Checking…
- Engine
- —
- Camera
- Off
- Model used
- —
- People in view
- —
- Threshold · target answer rate
- —
- Frames captured
- 0
- Frames processed
- 0
- Backlog
- 0
- Capture frames skipped
- 0
- Capture rate
- —
- Pose time p50 / p95
- —
- Queue age p50 / p95
- —
- Finish → text
- —
- Read aloud → speech start
- —
- Translation
- —
- Image features
- —
- Model score (lower is better)
- —
The score is a model diagnostic, not a percentage of accuracy. Each model has its own threshold, fitted on stored validation clips; thresholds do not transfer between models. Live-camera translation quality has not been validated. These details stay on this page and reset when you start the next utterance.