Research Assistant, Spoken Language System Group, MIT CSAIL
Developed an A.I. system that can detect depression in a conversation. From information on what was said (text) and how it was said (audio) the neural network model can automatically figure out what patterns it should look for. This is different to prior work, where a model is trained to look for specific words (`sad`, ‘down’, ‘therapy’) and questions (‘Do you have a history of depression’). She has also developed an A.I. system that can evaluate the mood of a conversation, by utilizing information form a speaker’s speech, pulse, and movement. The system works even without ever having heard the person before, and can determine to high accuracy whether they were happy or sad. Using a neural network, the system can automatically determine what signals to look for and can make evaluations in real-time.