Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.605887
Title: Advanced signal enhancement techniques with application to speech and hearing
Author: Chatlani, Navin
Awarding Body: University of Strathclyde
Current Institution: University of Strathclyde
Date of Award: 2011
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Advanced signal enhancement techniques with application to speech and hearing are presented that are applied to areas including adaptive noise cancellation (ANC) in noisy speech signals, single channel noise reduction for speech enhancement, voice activity detection (VAD) and noise reduction in binaural hearing aids. The performance enhancement of the new techniques over competing approaches is presented. For the domains of ANC and single channel noise reduction, the use of Empirical Mode Decomposition (EMD) is the underpinning technique employed. A novel approach to dual-channel speech enhancement using Adaptive Empirical Mode Decomposition (SEAEMD) is also presented, when a noise reference is available. The new SEAEMD system incorporates the multi-resolution approach EMD with ANC for effective speech enhancement in stationary and non-stationary noise environments. Two novel Empirical Mode Decomposition based filtering (EMDF) algorithms are presented for single channel speech enhancement. The first system is designed to be particularly effective in low frequency noise environments. The second generalized EMDF system is designed to operate under other noisy conditions, with results presented for babble noise, military vehicle noise and car interior noise. It is shown that the proposed EMDF techniques enhance the speech more effectively than current speech enhancement approaches that use effective noise estimation routines. Speech systems such as hearing aids require fast and computationally inexpensive signal processing technologies. A new and computationally efficient 1-dimensional local binary pattern (1-D LBP) signal processing procedure is designed and applied to (i) signal segmentation and (ii) the VAD problem. Both applications use the underlying features extracted from the 1-D LBP. The simplicity and low computational complexity of 1-D LBP processing are demonstrated. A novel binaural noise reduction system is presented for steering the focus direction of a hearing aid (2 microphones per hearing aid) to additional directions as well as 0/180 degrees. The system places a spatial null in the direction of the target speaker to obtain a noise estimate. The noisy speech signal is then filtered to perform noise reduction, and thus focus on the target speaker located at the desired direction. The results demonstrate its performance at attenuating multiple directional interferers.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.605887  DOI: Not available
Share: