Investigating bottom-up auditory attention

Abstract

Bottom-up attention is a sensory-driven selection mechanism that directs perception toward a subset of the stimulus that is considered salient, or attention-grabbing. Most studies of bottom-up auditory attention have adapted frameworks similar to visual attention models whereby local or global “contrast” is a central concept in defining salient elements in a scene. In the current study, we take a more fundamental approach to modeling auditory attention; providing the first examination of the space of auditory saliency spanning pitch, intensity and timbre; and shedding light on complex interactions among these features. Informed by psychoacoustic results, we develop a computational model of auditory saliency implementing a novel attentional framework, guided by processes hypothesized to take place in the auditory pathway. In particular, the model tests the hypothesis that perception tracks the evolution of sound events in a multidimensional feature space, and flags any deviation from background statistics as salient. Predictions from the model corroborate the relationship between bottom-up auditory attention and statistical inference, and argues for a potential role of predictive coding as mechanism for saliency detection in acoustic scenes.

Auditory salience using natural soundscapes

“…In particular, the other features seem helpful in establishing salient events in the context of acoustically dense scenes, in which the difference in loudness between the background and the salient sound is much lower. This finding is supported by Kaya et al 39 which showed that the contribution to salience from different features varied across sound classes. The Kaya et al study used comparatively more dense acoustic stimuli.…”

“…The second is a model by Kim et al, 19 which calculates salient periods using a linear salience filter and linear discriminant analysis on loudness. The third is a model by Kaya et al, 18 which detects deviants in acoustic features using a predictive tracking model using a Kalman filter. While the first two models are implemented as published in their original papers, the Kaya model is used with features expanded to include all the features included in the current study.…”

“…Most published studies used a detection paradigm where a salient object is embedded in a background scene, and subjects are asked to detect whether a salient event is present or not; or to compare two scenes and detect or match salience levels across the two scenes. [15][16][17][18] In these setups, salient events are a) Electronic mail: mounya@jhu.edu determined by an experimenter based on a clear distinction from the background scene. As such, their timing provides a ground truth against which accompanying models are tested, and the relevance of various features is assessed.…”

“…Of the few studies conducted on auditory salience, most have used wellcontrolled stimuli, like simple tones 22 or short natural recordings in controlled uniform backgrounds. 15,17,18 A few studies incorporated more rich natural scenes using recordings from the BU Radio News Corpus 23 and the AMI Corpus. 19 Still, both of the latter databases consist of scenes with relatively homogeneous compositions, both in terms of audio class (mostly speech interspersed with other sounds) and density (mostly sparse with rare instances of overlapping sound objects).…”

Prioritizing foreground selection of natural chirp sounds by tempo and spectral centroid

“…The former is a measure of the novelty of a sound "event", whereas the latter represents the quality driving the formation of the auditory foreground and background. We will focus on perceptual salience, complementing the literature on salience models that exclusively addresses sensory salience [5][6][7][8][9][10][11]. We equate perceptual salience with faster and more accurate stream selection.…”