Source localization from voice signals

Chen, Daguan

Permalink

https://hdl.handle.net/2142/105262 Copy

Description

Title

Source localization from voice signals

Author(s)

Chen, Daguan

Issue Date

2019-04-24

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

Choudhury, Romit Roy

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

Date of Ingest

2019-08-23T20:48:26Z

Keyword(s)

• Sound
• localization
• microphone array
• AoA
• sound source localization

Abstract

Indoor localization remains an open field of research, due to its utility and unresolved challenges. This thesis focuses on sound source localization using a small microphone array. Practically speaking, the aim is to allow smart voice assistants, such as Amazon Echo or Google Home, which possess a small array of microphones, to locate human speakers by determining the location of their voice in space. Although such voice assistants are currently capable of determining the azimuthal angle of arrival (AoA) of the human voice, they cannot determine the range of the speaker. This thesis addresses the two-dimensional localization problem: when a user speaks to an Amazon Echo, our goal is to be able to plot their location as a point on a bird’s eye view of the indoor floorplan. Our proposed solution, VoLoc, realizes two-dimensional sound source localization by determining the AoA of not only the direct path, but also one multipath. With two AoA directions, VoLoc is able to use inverse ray-tracing to find the sound source’s location in 2D space. One of the core challenges is to accurately distinguish the AoA of at least one multipath. To address the challenge, we introduce a new algorithm, iterative align-and-cancel. We observe median location accuracies of around 0.4m, across various real-world environments such as apartments, offices, and meeting rooms.

Graduation Semester

2019-05

Type of Resource

text

Permalink

http://hdl.handle.net/2142/105262

Copyright and License Information

Copyright 2019 Daguan Chen

Owning Collections