MODELING HUMAN SOUND-SOURCE LOCALIZATION WITH THE AUDITORY MODELING TOOLBOX

Spatial perception of sound sources involves different mechanisms in the auditory system, especially when it comes to localizing sound sources in the horizontal (left-right) or in the sagittal planes (front-back and top-down). Recently, a model for the perception of sound sources in the horizontal plane has been introduced (Takanen et al., 2014, Hear Res 309: 147163). It is based on the count-comparison analysis of binaural signals that is motivated by the spike-rate encoding of directional cues found in the mammalian auditory pathways. It results in an activation map visualizing the prediction of the perceived horizontal direction of sounds. Further, a model aiming at predicting the localization performance in sagittal planes has been introduced (Baumgartner et al., 2013, in The Technology of Binaural Listening, editor J. Blauert, [Springer, Berlin], Ch. 4, pp. 93-119). It is based on the comparison of processed acoustic signals with an internal template of processed head-related transfer functions and yields a statistical probability for the perceived direction of the source along a sagittal plane. These models can be extensively evaluated and applied because their implementation is publicly available, which is essential for reproducible research. The implementations of the models are available within the auditory modeling toolbox (AMT), an open-source and freely available collection of auditory models. In this contribution, we describe the AMT, the two models, and their implementations within the AMT, showing how to begin modeling human sound-source localization.