Title: Enhancing speech recognition performance in the presence of MRI scanner noise

Authors: Theologos Athanaselis, Stelios Bakamidis, Ioannis Dologlou, Stavroula-Evita Fotinea

Addresses: Institute for Language and Speech Processing (ILSP)/R.C. ''Athena'', Department of Voice and Sound Technology, Marousi, Athens, Greece. ' Institute for Language and Speech Processing (ILSP)/R.C. ''Athena'', Department of Voice and Sound Technology, Marousi, Athens, Greece. ' Institute for Language and Speech Processing (ILSP)/R.C. ''Athena'', Department of Voice and Sound Technology, Marousi, Athens, Greece. ' Institute for Language and Speech Processing (ILSP)/R.C. ''Athena'', Department of Voice and Sound Technology, Marousi, Athens, Greece

Abstract: In this paper, two well-known speech enhancement techniques are compared in a Magnetic Resonance Imaging (MRI) scanner noise reduction scheme prior to speech recognition experiment. Our study deals with the comparison between the Non Linear Spectral Subtraction (NSS) with iterative overestimation and the Singular Value Decomposition (SVD)-based noise reduction techniques in enhancing medical content speech contaminated by MRI scanner noise. It is proven experimentally that both techniques can improve the recognition performance of voice commands in order to voice control a MRI scanner. Making one step ahead, the paper further investigates the performance of both signal enhancement techniques in the recognition of speech utterances with medical information. The recognition results show that each technique improved the recognition accuracy as it was expected, but the NSS outperformed SVD.

Keywords: voice control MRI; speech recognition; NSS; nonlinear spectral subtraction; truncated SVD procedure; MRI scanning; MRI scanner noise; magnetic resonance imaging; medical content speech; signal enhancement.

DOI: 10.1504/IJSISE.2010.036890

International Journal of Signal and Imaging Systems Engineering, 2010 Vol.3 No.3, pp.179 - 187

Received: 19 May 2010
Accepted: 31 Aug 2010

Published online: 15 Nov 2010 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article