Subcortical Auditory Model including Efferent Dynamic Gain Control with Inputs from Cochlear Nucleus and Inferior Colliculus

Abstract

The focus of most existing auditory models is on the afferent system. The auditory efferent system contains descending projections from several levels of the auditory pathway, from the auditory cortex to the brainstem, that control gain in the cochlea. We developed a model with a time-varying, gain-control signal from the efferent system that includes sub-cortical ascending and descending neural pathways. The medial olivocochlear (MOC) efferent stage of the model receives excitatory projections from both fluctuation-sensitive neurons in the inferior colliculus (IC) and wide-dynamic-range neurons in the cochlear nucleus (CN). The response of the model MOC stage controlled cochlear gain dynamically. Changes in the rates of IC neurons in awake rabbit to long-duration amplitude-modulated (AM) noise were employed to adjust the parameters of the proposed model. In response to AM stimuli, physiological response rates of most IC neurons with band-enhanced (BE) modulation transfer functions (MTFs) increased over a time course consistent with the dynamics of the MOC efferent feedback. The time constant of the MOC model that best matched the IC physiology was compared to available descriptions of the MOC. Responses of the proposed subcortical model to AM noise simulate the trend of increasing rate over time, while the model without the efferent system did not show this trend.

Quantifying the Impact of Auditory Deafferentation on Speech Perception

“…Nevertheless, in this work, we showed that a neural network approach can be applied to decode speech from modeled ANF responses. This decoding method can be expanded to any of the already established models to study how perception would be affected by, for example, the loss of efferents in the auditory pathway ( Farhadi et al, 2022 ; Zilany et al, 2014 ).…”

“…For example, in the current model, we only considered the afferent system, while the efferent system also contributes significantly to hearing ( Cooper & Guinan, 2006 ; Guinan et al, 2003 ; Mukerji et al, 2010 ). Recent progress in modeling the peripheral efferent system has enabled the investigation of how its degradation affects speech perception in noise ( Farhadi et al, 2022 ; Grange et al, 2022 ). Moreover, with aging, a loss of afferent and efferent signals typically co-occurs ( Fu et al, 2010 ; Liberman & Liberman, 2019 ; Zhu et al, 2007 ).…”

A fast and flexible approximation of power-law adaptation for auditory computational models

“…Such simulations are prohibitively slow using true PLA, but parallel-exponential PLA is well suited to this usage. The latter is often encountered when working with high acoustic frequencies, when working with ideal-observer models that can be sensitive to the effects of aliasing (e.g., Heinz et al, 2001), or when simulating dynamic feedback loops that preclude intermediate downsampling steps to increase efficiency (Farhadi et al, 2023). Future work could explore better ways to fit the PLA approximation parameters and the optimal number of discrete processes to use in the approximation scheme to balance between computational costs and fidelity to true PLA.…”

Modeling neural coding in the auditory brain with high resolution and accuracy

“…A DNN architecture based on Conv-TasNet [49] was used to train an automatic-speech-recognition (ASR) back-end that predicted phonemes from the ICNet bottleneck response. The DNN architecture comprised: (1) a convolutional layer with 64 filters of size 3 and no bias, followed by a PReLU activation, (2) a normalization layer, followed by a convolutional layer with 128 filters of size 1, (3) a block of 8 dilated convolutional layers (dilation from 1 to 2 7 ) with 128 filters of size 3, including PReLU activations and residual skip connections in between, (4) a convolutional layer with 256 filters of size 1, followed by a sigmoid activation, (5) an output convolutional layer with 40 filters of size 3, followed by a softmax activation. All convolutional layers were 1-D and used a causal kernel with a stride of 1.…”

“…The brain, however, with its many interconnected and highly non-linear neural networks, is too complex for such an approach. Hand-designed models of the auditory brain can be effective in describing specific features of responses to a limited class of sounds [4][5][6][7], but more general models of the brain require the use of "black box" frameworks with parameters estimated from neural recordings. To facilitate model fitting from limited data, the allowable transformations are typically constrained to be either linear or low-order non-linear [2].…”