Adaptive Bayesian Learning with Action and State-Dependent Signal Variance