Acoustic Modeling: A Recurrent Research Field?

Oriol Vinyals

ICSI

Tuesday, February 28, 2012
12:30pm

This talk isn't going to be about a paper or a specific project, but rather an open talk about interdisciplinary topics in speech, vision, and machine learning. The main focus will be on acoustic modeling, which is an important component of speech recognition, and has been a constant topic of research for many decades. I'll give my view of current techniques and trends (many of which are not new per se, but rather a revisit of old work), followed by a brief introduction to state-of-the-art object recognition, which is another challenging machine learning problem in computer vision. I'll also report on preliminary experiments that I've been running exploiting synergies between both. By the end of the talk, I hope that some of you will be convinced about how speech and vision can (and do) learn from one another.