Computational framework for investigating predictive processing in auditory perception.

Area of Science:

• Auditory Neuroscience
• Computational Auditory Scene Analysis
• Perception and Cognition

Background:

• The brain predicts future sensory inputs by tracking sound sources over time.
• Previous predictive coding models primarily focus on predictable stimuli, limiting their application to complex, uncertain environments.

Purpose of the Study:

• To introduce a flexible computational model for investigating predictive coding in complex auditory scenes.
• To enable targeted questions about neural processes linking sensory input to behavior and neural responses.

Main Methods:

• A computational framework employing Bayesian inference to sequentially track sound features.
• Inference of sufficient statistics from past observations across multiple time scales for prediction.
• Tracking the statistical structure of sensory input to model auditory perception.

Main Results:

• Demonstration of the model's flexibility in capturing diverse statistical structures in real-world audio.
• Application of model facets to perceptual research, illustrating its utility.

Conclusions:

• The model serves as a general framework for hypothesis generation in predictive coding research.
• It guides investigations into neural mechanisms underlying the perception of complex auditory scenes.