Automating acoustic signal processing experiments and audio machine learning datasets using robots

Lu, Austin

Permalink

https://hdl.handle.net/2142/127217 Copy

Description

Title

Automating acoustic signal processing experiments and audio machine learning datasets using robots

Author(s)

Lu, Austin

Issue Date

2024-12-11

Director of Research (if dissertation) or Advisor (if thesis)

• Singer, Andrew C
• Corey, Ryan M

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Audio signal processing
• Microphone arrays
• Spatial audio
• Robotics
• 3D printing

Abstract

We develop specialized robots for audio and acoustic experiments, which in turn facilitate acoustic signal processing and audio machine learning. Our robo-acoustic mannequin, a low-cost 3D printed device, is custom-made to enable interesting spatially-dynamic experiments. We explore simple solutions to quiet actuation, thus avoiding the infamous problem of audible robot noise, and we open-source our design to stimulate further development. Using this new research resource, we empirically study how motion, specifically head-turning, affects the objective performance of a spatially-adaptive MVDR beamformer. We find the surprising result that the presence of motion has a significant effect, but the rate of motion does not. To study more intricate scenarios, we also design a multi-robot mechatronic recording studio that automatically captures high-resolution labeled audio datasets. Large-scale multiple-day-spanning recordings that would be impossible to do manually become possible through this work. The multi-robot system is accessed by geographically disparate researchers via a “DAW for robots” web interface, thus demonstrating the potential of shared, collaborative audio-robot workspaces. Overall, we expect our work to motivate new and interesting robot-enhanced audio experiments and datasets.

Graduation Semester

2024-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/127217

Copyright and License Information

Copyright 2024 Austin Lu

Owning Collections