CLERC · THE DATA LAYER FOR SIGN LANGUAGE AI
Everyone chases the model.
The data is the bottleneck.
6
Native Deaf signers
1,200
ASL videos
71 min
Signed duration
201
Parallel phrases
128
Keypoints / frame
Open
On Hugging Face
epee
1,200 clips · 71 min · 6 native Deaf signers · 201 parallel phrases
See the dataset in action: try the live ASL sign recognition demo — it runs in your browser on a model trained with this data.
LIVE / POSE KEYPOINTS
[ ALPHA ]
[ BRAVO ]
[ CHARLIE ]
[ DELTA ]
[ ECHO ]
All six Deaf signers · real motion keypoints · one corpus. The five on the parallel grid sign a shared phrase; FOXTROT signs his own, from the second register.
WHAT THE DATA PROVES · PICK ONE
READ THE RESEARCH
ESSAY
Why Data, Not the Model
Sign language AI does not have a model problem. It has a data problem, and that changes everything.
TECHNICAL
From Raw Video to Labeled Signs
How CLERC transforms raw footage into production-ready, millisecond-segmented sign language data.
DEMO
The CLERC Demo
100 videos. 1,100 annotated glosses. 90% recognition accuracy. The pipeline, end to end.