Spiking neural networks for auditory spatial perception: decoding sound direction from auditory neural representations

Human sound localization relies on spatial information inferred by the auditory system. In the present work, two psychophysically informed models are proposed, in which perceived sound direction is inferred directly from physiologically grounded auditory neural representations of the acoustic signals arriving at the ears. The auditory nerve fiber (ANF) and medial superior olive (MSO) firing rate features are generated by a spiking auditory front-end, and continuous lateral and circumaural angles are estimated by feedforward neural networks. Lateral angles are estimated from either ANF or MSO firing rates, whereas circumaural angles are estimated from positive spectral gradients (PSGs) of ANF firing rates. The models are designed to reproduce human sound localization behavior while preserving physiologically meaningful intermediate representations. Interpretability is therefore only partially retained: although the decoder itself is not transparent, the intermediate representations remain physiologically structured and can be systematically manipulated. Across the tested training strategies, behavioral-response labels were treated as statistical abstractions of stochastic human localization data. Average-response labels provided the most consistent agreement with subject-specific behavioral trends, whereas Best-response labels, evaluated against the corresponding best-response behavioral reference, did not consistently reduce model–human discrepancy. Task-level decoding accuracy and perceptual fidelity also ranked the auditory representations differently: ANF-based decoding achieved higher physical localization accuracy under some acoustic conditions, whereas the MSO-based pathway better reproduced characteristic human lateral-error patterns. The results show that flexible neural decoders can reproduce key features of human spatial perception while maintaining a link between auditory coding and behavior.