Sensorineural hearing loss (SNHL) impairs auditory perception, reflecting degraded representations in the central auditory system. While studies using simple stimuli (e.g., tones, noise) have revealed cortical changes consistent with behavioral deficits, it remains unclear how these changes affect the encoding of real-world sounds. We recorded from 4000 neurons in the auditory cortex of awake ferrets with either normal hearing or mild-to-moderate noise-induced hearing loss (NIHL), induced using 2.8–5.6 kHz noise (116 dB SPL, 2 h). Neural responses to 50 natural sound categories (ESC-50) were classified using linear spike-count-based population decoders. Accuracy reached 50% in normal-hearing animals, but was substantially reduced following NIHL. To probe tuning changes, we trained convolutional neural networks (CNNs) to predict neural responses to 30 h of natural sounds. These models explained over 50% of the explainable variance in neural responses and identified low-dimensional cortical encoding manifolds for each animal. Following hearing loss, manifold dimensionality was reduced, and key dynamic properties, such as spectrotemporal tuning and contrast gain control, were altered. These results provide new insight into how NIHL degrades cortical sound representations and how encoding manifolds capture the underlying plasticity-related changes. Work supported by NIH K99DC022330 (SP), R01DC014950 (SD).
Parida et al. (Wed,) studied this question.