The ability to listen and identify sounds in the presence of competing background noise is a critical function of the healthy auditory system. Humans with normal hearing can easily carry a conversation even with relatively high levels of noise and in complex auditory environments, such as a busy restaurant. Yet, for individuals with hearing loss even moderate levels of background noise can adversely impact sound recognition. Understanding the neural mechanisms that underlie recognition in noise is thus of high clinically relevance. This project provides a novel approach to study how the healthy auditory system utilizes statistical sound cues during real-world sound recognition in the presence of competing natural background sounds. Using human participants listening to natural sound mixtures and synthetically modified variants, Aim 1 explores how several sound texture statistics influence the perception of speech in real-world noise. Aim 2 tests the hypothesis that the same statistical sound cues modulate neural activity in auditory midbrain and that these statistics influence, beneficially or detrimentally, the neural representation of a foreground sounds in natural noises. Neural decoders will then assess how the neural activity contributes towards recognition under various natural noises with distinct statistics. Finally, in Aim 3, a model that captures peripheral and central auditory system transformations will be used to mechanistically predict neural activity, neural decoding performance, and human recognition, under adverse natural masking conditions. We hypothesize that, by capturing the fundamental transformations of the central auditory system, we will be able to predict neural-based recognition and human perceptual trends. The study will lay a foundation for developing a general theory of how the auditory system utilizes high-order statistical structure for natural, realistic sound recognition in background environmental noise. The results will extend previous work on auditory masking by quantify the influence of high-order cues on perception, will provide detailed descriptions of how statistical cues drive non-classical neural responses, and will define models that can account for both physiology and behavior. The findings, models, and optimal quality metrics that will be developed have health related implications that can potentially improve human communication outcomes, with likely applications for hearing diagnosis and developing biologically inspired noise suppression strategies for auditory prosthetics.