Model-Based Speech Enhancement

Harding, Philip (2013) Model-Based Speech Enhancement. Doctoral thesis, University of East Anglia.

[thumbnail of Philip-Harding-PhD-2013.pdf]
Download (15MB) | Preview


A method of speech enhancement is developed that reconstructs clean speech from
a set of acoustic features using a harmonic plus noise model of speech. This is a significant
departure from traditional filtering-based methods of speech enhancement.
A major challenge with this approach is to estimate accurately the acoustic features
(voicing, fundamental frequency, spectral envelope and phase) from noisy speech.
This is achieved using maximum a-posteriori (MAP) estimation methods that operate
on the noisy speech. In each case a prior model of the relationship between the
noisy speech features and the estimated acoustic feature is required. These models
are approximated using speaker-independent GMMs of the clean speech features
that are adapted to speaker-dependent models using MAP adaptation and for noise
using the Unscented Transform.
Objective results are presented to optimise the proposed system and a set of subjective
tests compare the approach with traditional enhancement methods. Threeway
listening tests examining signal quality, background noise intrusiveness and
overall quality show the proposed system to be highly robust to noise, performing
significantly better than conventional methods of enhancement in terms of background
noise intrusiveness. However, the proposed method is shown to reduce signal
quality, with overall quality measured to be roughly equivalent to that of the Wiener

Item Type: Thesis (Doctoral)
Faculty \ School: Faculty of Science > School of Computing Sciences
Depositing User: Users 2259 not found.
Date Deposited: 05 Mar 2014 11:14
Last Modified: 05 Mar 2014 11:14

Actions (login required)

View Item View Item