MACHINE LEARNING ENABLED TRANSLATIONS
THE ARCHITECTURE
Using groundbreaking pre-processing technology, Universal Messaging is able to understand the meaning of various signs.
By utilizing hand landmarks, we receive data that is then efficiently processed and interpreted by a Machine Learning model.
REAL-TIME TRANSLATION
By pre-processing our video data, Universal Messaging is able to deliver sign language inferences with industry-leading speeds on consumer devices. Each sign can be interpreted within 32 ms on a 2019 Apple MacBook Pro, with a validation accuracy of 96.13%.
LOW STORAGE & HIGH EFFICIENCY
By taking hand landmarks and normalizing the coordinates, we perform all computations on a user's device. Less processing power is required to interpret a frame, while the data used for inference in our model is stored as an array of 42 floating point numbers.
HOW CONNECTIONS WORK
We’ve begun the development of our website, which will allow users to communicate through sign language with others who may not understand ASL.
ADVANCED CYBER SECURITY
End-to-end encryption between users enables secure communication of messages. Alongside this, the ability to interpret signs in real-time on a user's device allows us to keep all data with the user - none of it ever leaves their possession.