Sound localization in audio-based games for visually impaired children

Abstract

This thesis describes the design of a sound localization algorithm in audio-based games for visually impaired children. An algorithm was developed to allow real-time audio playback and manipulation, using overlapping audio blocks multiplied by a window function. The audio signal was played through headphones. Multiple sound localization cues are evaluated. Based on these cues, two basic sound localization algorithms are implemented. The first uses only binaural cues. The second expands this with spectral cues by using the head-related impulse response (HRIR). The HRIR was taken from a database, interpolated to obtain optimal resolution and truncated to minimize memory usage. Both localization algorithms work well for lateral localization, but front-back confusions using the HRIR are still common. The signal received from a sound source changes with the distance to the sound source. Both the distance attenuation and propagation delay are implemented. As an alternative means of resolving front-back ambiguities, the use of a head tracker was investigated. Experiments with a webcam based head tracker show that a head tracker is a powerful tool to resolve front- back confusions. However, the latency of webcam based head trackers is too high to be of practical use.