Multi-Speaker Set-up for Audiological Measurements using Gesture Recognition

Spatial hearing measurements conducted under free sound-field conditions require a pointing method which allows the subject to indicate the perceived auditory event. Routine investigations with traditional set-ups often include interactions between the operator and the subject, leading to time-consuming test procedures with complex instructions. This thesis presents a multi-speaker set-up with a new approach to pointing methods using gesture recognition. An audio processing framework for the software of the measurement set-up (AIODE) was implemented to provide the hardware controls and the signal processing routines required for sound localization and speech intelligibility experiments in the horizontal plane. Based on room acoustics measurements and the specifications given by EN ISO 8253-2 and EN ISO 8253-3, the test environment has been characterized. The linear behaviour and acoustic output of the system were validated and a calibration procedure was elaborated. The set-up provides acoustic stimuli with up to 90 dB SPL and a flat frequency response between 0.7 and 12.5 kHz. The pointer method using automatic gesture recognition was realized with a MICROSOFT KINECT sensor. The temporal tracking characteristics and the detection accuracy of the system were analyzed and evaluated. The presented set-up supports the automated detection of indicated directions with 15° accuracy. A practicability test with five volunteers with relation to the institute was conducted, investigating effects on the responses through visual cues given by the speakers. After a short training session (<5 min), test procedures with 36 test steps were performed in less than 5 min. The experiences made in the practicability test promise good applicability of the measurement set-up with gesture recognition for further audiological experiments.