Neural-scaled entropy as a model of information for speech perception

Abstract

Neural-Scaled Entropy (NSE) is an objective metric used to quantify “information” available in speech consequent hearing loss, hearing aid signal processing, and distortion from various environmental factors. One pursuit is to use NSE to find optimum hearing aid settings that maximize speech perception. Inspired by the Cochlear-Scaled Entropy model [Stilp et al., J. Acoust. Soc. Am. 2112–2126 (2010)], NSE uses the neural spike output at the inner hair cell synapse of an auditory nerve model [Zilany et al., J. Acoust. Soc. Am. 126, 2390–2412 (2009)]. Probability of spike output from fibers sampled at equidistant places along the model cochlea is computed for short duration time frames. Potential information is estimated by using the Kullback-Liebler Divergence to describe how the pattern of neural firing at each frame differs from preceding frames in an auto-regressive manner. NSE was tested using nonsense syllables from various perceptual studies that included different signal processing schemes and was compared to performance for different vowel-defining parameters, consonant features, and talker gender. NSE has potential to serve as a model predictor of speech perception, and to capture the effects of sensorineural hearing loss beyond simple filter broadening. [Work supported by NIDCD RC1DC010601.]