Automatic Recognition of Acoustic Scenes

Dan Ellis

Columbia University, New York

Tuesday, August 4, 2015
12:30 p.m., Conference Room 5A

Humans and other animals have developed a sense of hearing because of the rich, complementary information available from the sound environment.  Most work on automatic sound processing has focused on the specific (and important) case of speech, but in this talk I will consider more general environmental sounds - the ones that led to the evolution of hearing in the first place.  Automatic recognition of acoustic environments has applications in multimedia archive management, in context sensitivity for smart devices, and in auditory prostheses, among many other applications.  I will discuss current and future developments in the features, classifiers, and data sources used for this technology, moving towards the goal of general-purpose automatic labeling of soundtrack events.

Bio:

Dan Ellis is a Professor in the Electrical Engineering Department of Columbia University in New York, where he leads the Laboratory for Recognition and Organization of Speech and Audio (LabROSA).  Prior to joining Columbia in 2000, he was a Research Scientist at the International Computer Science Institute in Berkeley, CA.  His Ph.D. is from the MIT, where he was a Research Assistant at the Media Lab.  His research interests center around applying machine learning and signal processing to extract information from real-world sound signals, including speech, music, and environmental sounds.  For a sabbatical starting in September 2014, he joined Google's Sound Understanding Group where he has been working on the Daredevil project to develop general-purpose audio event detectors.