Learn
Practise against recordings of real signers. Every attempt is scored with corrections down to individual fingers.
Open Learn →Suvana
SUVANA
සුවණ
One platform for two-way communication between Deaf and hearing people, in Sri Lankan Sign Language.
Turn signs into speech · Speak to a signing avatar · Learn to sign · Stay aware of your surroundings
Getting started
A language of hands, faces and movement.
Sri Lankan Sign Language is a full natural language — grammar carried in handshape, orientation, motion and facial expression, not a transliteration of Sinhala. Around 400,000 Deaf and hard-of-hearing Sri Lankans use it, and almost no hearing services do.
Suvana works in both directions at once: it reads signing and speaks it, speaks back in signing, teaches hearing people to sign, and keeps Deaf users aware of sound they cannot hear.
Practise
Sign to the camera in Recognize and watch the confidence bar climb until the model reads you cleanly.
Find Recognize →Converse
Speak Sinhala and watch a 3D avatar sign it back — the round trip that makes a two-way conversation possible.
Open Communicate →
—
Signs recognised
4
Integrated modules
75
3D landmarks / frame
98.8%
Sequence test accuracy
≤300ms
Feedback latency target
Modules
One platform,
four bridges.
Sign → Speech
Recognize
Sign to the camera and have it recognised and spoken aloud, built on a spatial MediaPipe Holistic and temporal sequence model.
SIGN → SPEECH
Speech → Sign
Communicate
Speak Sinhala and watch a 3D avatar sign it, with the emotion in your voice detected along the way.
Open Communicate
Live
For hearing learners
Learn
Practice Sri Lankan Sign Language with your camera. Every attempt is scored against real-signer references, with corrective feedback down to individual fingers.
Open Learn
Live
Sound awareness
Alerts
A companion phone app that keeps Deaf users aware of important sounds around them and can send SOS alerts.
Open Alerts
Companion phone app
Why it works
Research decisions,
not feature checkboxes.
Real-time, in the browser
Landmark extraction runs on the device with MediaPipe, so practice needs no install and camera frames never have to leave the machine to be tracked.
Scored against real signers
Attempts are compared with dynamic time warping against reference recordings, which scores the whole movement — handshape, orientation and trajectory — rather than a single frozen frame.
Corrections, not just a score
Per-joint and per-frame deviation is what the comparison actually produces, so feedback can say which finger drifted and where in the movement it happened.
Practice that follows the learner
Every attempt updates a running estimate of how well each sign is known, weighting the next session toward the weakest and least recently drilled signs.
Architecture & AI Pipeline
Bridging Vision, Sound & Avatars
Suvana coordinates four specialized deep learning and spatial computing systems to power full-duplex communication in Sri Lankan Sign Language.
01
Spatial Computer Vision
Extracts 75 3D keypoints across hands, face contours, and pose landmarks in real time using MediaPipe Holistic at 60 FPS.
On-Device Edge
02
Temporal Sequence AI
Processes continuous 30-frame temporal gesture windows with deep LSTM and Transformer sequence models to decode dynamic signs.
Sequence Classifier
03
Sinhala TTS & 3D Avatar
Instantly speaks Sinhala translations with warmed audio cache, and synthesizes 3D signing avatars with vocal emotion detection.
Speech & Graphics
04
Acoustic Awareness
Mobile edge classifier detects environmental hazard sounds like sirens, alarms, and horns, sending haptic and flash alerts.
AlertsEnd-to-End Pipeline
How Sign Recognition Works
From raw camera frames to synthesized Sinhala speech: our pipeline isolates signer kinematics, normalizes 3D spatial coordinate tensors, classifies sequence patterns, and outputs speech synthesis with zero server storage.
MediaPipe Holistic
LSTM Sequence Model
On-Device Inference
Sinhala Speech Engine
Community & Feedback
Tested by designers, developers, & advocates.
The sign language AI web app is an innovative and user-friendly tool that accurately detects hand gestures and translates them into corresponding signs. It ensures reliable performance, making it an excellent resource for communication and education.
The user experience of the application is very smooth and easy to understand. Used this application to try out different signs and detection was smooth and prompt.
The advancement in the field of sign language is really helpful for the deaf and hard of hearing community — it eliminates the need of a middle person by just using the website. The UI is super smooth and user-friendly.
· 4 Integrated Research Modules
· 75 3D Landmarks per Frame
· 98% Sequence Accuracy
· Dual-Script Sinhala & English