Predicting the location of expected upcoming stimuli is a key function in sensory processing. Noise in the environment reduces the reliability of sensory input and increases the importance of prior information. Yet, sudden changes in the environment can render this prior information irrelevant. In such cases, observers should reset their beliefs and establish new priors using current sensory evidence. To study these dynamics of perceptual belief updating, we conducted a human EEG experiment (N = 29) investigating auditory spatial localization in noisy volatile environments and analyzed the associated behavioral and neural patterns using a Bayesian inference framework.Participants listened to sequences of sounds, in which each sound had a 1/6 chance of being a change point at which the sound location changed. At the end of each sequence, they indicated the location of the final sound. As a different number of sounds was presented in each trial, participants were expected to constantly update their spatial estimations.Using a computational model, we have regressed event-related potentials and neural oscillations in theta, alpha, and beta range against latent model variables reflecting prior and likelihood parameters for every individual sound of the sequences.Our results indicate, that the neural response to each sound reflects a reliability-based weighting between prior and likelihood, and supports the notion of Bayesian-like auditory inference in dynamic environments.