Investigating bottom-up auditory attention.

Area of Science:

• Auditory Neuroscience
• Computational Auditory Scene Analysis

Background:

• Bottom-up attention selects salient stimuli, often defined by contrast in visual models.
• Auditory attention research has largely adapted visual attention frameworks, focusing on contrast-based saliency.

Purpose of the Study:

• To develop a novel computational model of auditory saliency.
• To examine the multidimensional space of auditory saliency, including pitch, intensity, and timbre.
• To investigate the interactions among auditory features in saliency detection.

Main Methods:

• Developed a computational model based on psychoacoustic findings and auditory pathway processes.
• Modeled perception as tracking sound event evolution in a multidimensional feature space.
• Hypothesized that deviations from background statistical regularities are flagged as salient.

Main Results:

• The model's predictions align with the link between bottom-up auditory attention and statistical inference.
• Demonstrated the role of feature interactions in auditory saliency.
• Provided evidence supporting predictive coding in acoustic saliency detection.

Conclusions:

• Auditory saliency is fundamentally linked to statistical inference and deviations from expected sound patterns.
• Predictive coding may serve as a key mechanism for detecting salient events in complex acoustic environments.
• This work offers a new framework for understanding bottom-up auditory attention beyond simple contrast models.