A dynamic computational auditory scene analysis model of hearing using artificial neural networks

Abstract

Of note to this document will be the utility of neural networks in computational binaural modeling.

Humans perform various psychoacoustical phenomena when presented with a stimulus in a real room in the presence of reflections. Specifically, for a given sound source, humans are able to perceive directional cues contained within both the direct sound and reflections, and can internally resolve this information. This is known as the precedence effect. While various phsychoacoustical phenomena with different properties fall under this term's use, this document will specifically focus on the properties of reflections that occur within a short time window of the direct sound, how the human auditory system is able to resolve such content, how a computational model can be programmed to identify the binaural cues within these reflections, and how a computer can resolve the content of these reflections.

The human auditory system includes highly complex and robust methods of directional processing and filtering in order to extract signals from real environments. Humans rely on the auditory system for communicating, listening for potential danger, and appreciating and performing of music. Using only two ears as receivers, the auditory system can extract a wide range of information from sound in physical environments. This includes the ability to parse mixed, concurrent acoustical scenes into individual streams as well as the ability to resolve competing directional information contained within room reflections. Acoustical scenes and streaming are studied under a subset of the field of acoustics known as Auditory Scene Analysis, and studies demonstrate that the ability may be reinforced by head movement. Many of the directional cues incorporated in both the earliest-arriving wavefront -- known as the direct sound -- and from delayed reflections are resolved in the auditory system, and used to reinforce the content within the original stimulus. While only some of the auditory processing that allows humans to perform these functions is understood, there exists an extensive body of research offering insight into human performance in complex listening environments.