Flexible processing of Complex Auditory Scenes using Attention

Abstract

Analyzing constant streams of big data arriving from sensor arrays in an energy efficient manner is a formidable challenge for artificial systems. The brain tackles this problem in an elegant manner using the mechanism of attention. The goal of this project is to design a brain inspired audio processing algorithm that leverages attention to process audio streams from multiple spatial locations in a selective, flexible, and energy efficient manner. Specifically, this project will integrate a biologically based bottom-up algorithm for sound segregation (BOSSA), with a model of auditory attention (AIM), to develop a powerful top-down algorithm for audio processing in real-world noisy scenes with multiple sound sources. Compared to current alternatives, e.g., beamformers and deep neural networks, this algorithm will enable implementations that are more energy efficient and have a compact form factor. The algorithm will transform audio processing in complex scenes with multiple auditory sources, a challenging condition for both humans, e.g., those with hearing impairment, ADHD, or autism; as well as artificial speech recognition algorithms (e.g., SIRI or Alexa). Potential defense applications include audio surveillance and human machine communication in noisy scenes.

Document Details

Document Type
DoD Grant Award
Publication Date
Nov 08, 2024
Source ID
N000142412449

Entities

People

  • Kamal Sen

Organizations

  • Boston University
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science
  • Engineering

Readers

  • Distributed Systems and Data Platform Development
  • Neural Network Machine Learning.
  • Speech Processing/Speech Recognition.

Technology Areas

  • AI & ML