Back to list

Modeling Sound Localization in Sagittal Planes for Human Listeners

Authors Baumgartner, R.
Year 2015
Thesis Type Doctoral thesis
Topic Psychoacoustics
Abstract Sound localization in sagittal planes includes estimating the source position in up-down and front-back direction, and is based on processing monaural spectral cues. These cues are caused by the directional, acoustic filtering of the sound by the listener’s morphology, as described by head-related transfer functions (HRTFs). In order to better understand the listener-specific localization process and to reduce the need for costly psychoacoustic experiments, this PhD project aims at developing a functional model for sound localization in sagittal planes. In the model, directional spectral cues of incoming sounds are compared with internal templates of cues in order to obtain a probabilistic prediction of the listener’s directional response and localization performance. As directional cues, the model extracts rising spectral edges from the incoming sound spectrum. The comparison of cues is further influenced by the listener-specific sensitivity, representing the ability to process spectral cues. After extensively evaluating the model’s predictive power with respect to effects of HRTF or source-signal modifications on localization performance, particular model components were undertaken hypothesis-driven analyses. Model predictions were used to explain whether the large across-listener variability in localization performance is more attributed to listener-specific HRTFs, a purely acoustic factor, or to listener-specific sensitivities, a purely nonacoustic factor. The results of a systematic permutation of the factors suggest that the non-acoustic factor influences listener-specific localization performance much more than the acoustic factor. In order to investigate the effect of extracting spectral edges as directional cues, model predictions with and without edge extraction were compared. Predictions with edge extraction showed a better correspondence with results from localization experiments and explain the robustness of spectral cues in face of macroscopic variations of the source spectrum. The potential of the model to assess the spatial sound quality of technical devices was addressed. In particular, the microphone position in hearing-assistive devices and the positioning of loudspeakers in surround sound systems applying vector-based amplitude panning was investigated. For the sake of usability and reproducibility, the implementation of the model and the simulated experiments were made publicly available at the Auditory Modeling Toolbox.
Supervisors Höldrich, R., Eckel, G., Majdak, P., van de Par, S.