Computer Vision News - November 2020

2 Young Scientist Winner 0 Best of MICCAI 2020 This work is part of a bigger NSERC- funded project called Brain2Speech , led by co-authors Sidney Fels and Bryan Gic k, which is trying to understand the underlying mechanism behind speech from the formation of active parts in the brain. The project is divided into three parts: brain to muscle, muscle to articulation , and this part, articulation to speech . Thinking about next steps, Pramit says he is motivated by the opportunity to delve deeper into how speech motor control works and why we find speech tasks so easy to perform without considerable effort, despite speech being one of the most complex human processes requiring precise movements of multiple muscles. “We are trying to get an information- theoretic view of speech motor control,” Pramit tells us. “In other words, how difficult is one speech task with respect to another speech task? For example, how difficult is saying ‘how are you?’ rather than ‘hi’? For difficult tasks, the effort level is much higher, which means that more information is demanded from the motor control part. Now, because we have the mapping between the tongue movements and the formant frequencies ready, we can run experiments to investigate the difficultly level of different tongue movements and see if it reflects the same difficulty level in the acoustic space.”