Autistic children and non-verbal children alike face many issues in communicating their emotions, needs, requests, or thoughts. This project aims to understand how machine learning can be applied to classify the vocal cues of a non-verbal child and be integrated into a small device to be used as a communication tool (planned to be Raspberry Pi).
The model is currently trained on the ReCANVo dataset by Dr. Kristy Johnson and Dr. Jaya Narain. ReCANVo was used for a similar use case in the COMMALLA project at MIT Media Lab, aiming to classify non-verbal vocal cues. This project will take different approaches to see how the model can be 1) improved and 2) deployed onto small devices for classroom setting usage.
All ReCANVo audio is available here.