Spiking network optimized for word recognition in noise predicts auditory system hierarchy

Area of Science:

• Neuroscience
• Computational Neuroscience
• Auditory Processing

Background:

• The auditory system effectively recognizes speech amidst acoustic variability and background noise.
• The neural transformations underlying this noise resilience remain largely uncharacterized.

Purpose of the Study:

• To investigate the neural computations enabling robust auditory perception in noisy environments.
• To model the organizational hierarchy of the ascending auditory pathway using a computational approach.

Main Methods:

• Development and optimization of a hierarchical spiking neural network (HSNN) for word recognition accuracy in noise.
• Comparison of HSNN organizational principles with neurophysiological data from the auditory nerve, midbrain, thalamus, and cortex.
• Evaluation of network performance based on information transfer versus task-specific optimization.

Main Results:

• The optimized HSNN successfully predicted the hierarchical organization of the auditory pathway, including decreased temporal resolution and increased sparseness.
• The network's optimal scheme selectively filtered noise and non-essential temporal cues, enhancing word recognition.
• Networks optimized for high information transfer or conventional feature extraction models showed poorer performance and failed to predict auditory pathway organization.

Conclusions:

• The hierarchical structure and sequential nonlinear computations in the auditory pathway are crucial for enhancing relevant acoustic cues and suppressing noise.
• This organizational scheme optimizes sound representation, particularly in noise-compromised conditions.
• Computational modeling provides insights into the functional principles of auditory processing and neural coding.