Kurtosis-based blind beamforming: an adaptive, subband implementation with a convergence improvement

In many speech applications, a single talker is captured in the presence of background noise using a multi-microphone array. Without knowledge of the array geometry, talker location, or the room response, many traditional beamforming techniques cannot be used effectively. An adaptive, maximum-kurtosis objective is used in the frequency domain to blindly enhance the speech signal. The algorithm provides SNR gains of 3.5 - 7.5 dB with just two microphones in low-SNR, real-world scenarios. An improvement is presented that allows for faster and more stable convergence of the algorithm in real-time implementations. Finally, an alternative formulation to the problem is given, framing it in a way that might inspire new discussion or alternative solutions.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical and Computer Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2014

Author dc:creator

• Klingler, Daniel

Contributors dc:contributor

• Jones, Douglas L.

• Speech Enhancement
• Maximum Kurtosis
• Subband Implementation
• Convergence Improvement
• Beamforming
• Noise Reduction

Statement dc:rights

• Copyright 2013 Daniel Klingler

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/46649

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/46649