Incorporating softmax in psychophysical detection models for normal and electric hearing

Abstract

Modeling psychophysical auditory detection has proven to be difficult, as with existing neural models and detection models, we were unable to adjust the slope of the psychometric curve accurately. In machine learning, the softmax function is an excellent tool to assign probabilities to model outputs. Incorporating this function into psychophysical detection models can enhance the precision of the auditory detection model. This study extended Hamacher's detection model by integrating a softmax function, providing additional control over the slope of psychometric curves. center dot Using computational simulations of both normal and electric hearing, we applied this enhanced model to two psychophysical tasks: masker-probe detection and amplitude modulation detection. center dot The outcomes demonstrated that the normal hearing model aligned closely with expected performance, with predictable shifts in psychometric curves as the noise and slope parameters varied. In addition, with the electric hearing model, the new detection model could now reach lower asymptotes in the psychometric curve than with Hamacher's detection model. center dot These findings suggest that incorporating the softmax function provides a flexible tool for modeling auditory psychophysical data. This tool has potential applications for in silico evaluation of speech coding strategies for cochlear implants.