Robust Acoustics And Speech Perception Of Aerial Robot Under Ego Noise For Scene Understanding During Critical Emergency Response Missions
Abstract
For aerial robot based robust scene understanding, it is required that each sensor on the aerial robot acquire reliable sensor data. Due to the flight condition of the aerial robot and the ensuing emergency situation mission, however, the robot platform suffers a large uncertainty of data consistency and differences with normal ground condition. Information bearing acoustic sources such as human speech and other acoustic events representing abnormal situation particularly represents such case as the airborne platform encounters noise bearing sources convoluted with wind noise. Therefore, it is necessary to investigate novel signal processing algorithms for acoustic noise reduction and signal enhancement under harsh environments such as ego/wind/propeller fan noise of aerial robot so that speech event can be processed for time critical scene understanding. Auditory scene analysis in aerial robots is technically extremely challenging due to the inherent issues. For example, high level of ego-noise from rotor convoluted with wind, constantly changing noise level and target to sensor distance while robot is moving, and wide dynamic range of target signal power by changing target to sensor distance in outdoor environment. We plan to address these challenges in aerial robot platform and achieve robust auditory scene analysis by developing robust target sound source localization in highly noisy environment, effective multichannel signal enhancement to separate each acoustic event from mixed inputs based on spatial information, and reliable audio event recognition of highly distorted features. While new techniques for auditory scene analysis have been continuously studied and introduced for more than a decade, main focus has been on indoor environment for human-robot interaction, and very few work has been done on aerial robot perception.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 05, 2016
- Source ID
- FA23861614130
Entities
People
- Hanseok Ko
Organizations
- Air Force Office of Scientific Research
- Korea University
- United States Air Force