eduzhai > Applied Sciences > Engineering >

C-SL Contrastive Sound Localization with Inertial-Acoustic Sensors

  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Human brain employs perceptual information about the head and eye movementsto update the spatial relationship between the individual and the surroundingenvironment. Based on this cognitive process known as spatial updating, weintroduce contrastive sound localization (C-SL) with mobile inertial-acousticsensor arrays of arbitrary geometry. C-SL uses unlabeled multi-channel audiorecordings and inertial measurement unit (IMU) readings collected during freerotational movements of the array to learn mappings from acousticalmeasurements to an array-centered direction-of-arrival (DOA) in aself-supervised manner. Contrary to conventional DOA estimation methods thatrequire the knowledge of either the array geometry or source locations in thecalibration stage, C-SL is agnostic to both, and can be trained on datacollected in minimally constrained settings. To achieve this capability, ourproposed method utilizes a customized contrastive loss measuring the spatialcontrast between source locations predicted for disjoint segments of the inputto jointly update estimated DOAs and the acoustic-spatial mapping in lineartime. We provide quantitative and qualitative evaluations of C-SL comparing itsperformance with baseline DOA estimation methods in a wide range of conditions.We believe the relaxed calibration process offered by C-SL paves the way towardtruly personalized augmented hearing applications.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...