Hu_BSPC-D-20-00930_R2.pdf (1.71 MB)
Download file

A study of decodable breathing patterns for augmentative and alternative communication

Download (1.71 MB)
journal contribution
posted on 08.01.2021, 10:41 by Yasmin Elsahar, Sijung HuSijung Hu, Kaddour Bouazza-MaroufKaddour Bouazza-Marouf, David KerrDavid Kerr, Will Wade, Paul Hewett, Atul Gaur, Vipul Kaushik
People who use high-tech augmentative and alternative communication (AAC) solutions still face restrictions in terms of practical utilization of present AAC devices, especially when speech impairment is compounded with motor disabilities. This study aims to explore an effective way to decode breathing patterns for AAC by the means of a breath activated dynamic air pressure detection system (DAPDS) and supervised machine learning (ML). The aim is to detect a user’s modulated breathing patterns (MBPs) and turn them into synthesized messages for
conversation with the outside world. MBPs are processed using a one-nearest neighbor (1-NN) algorithm with variations of dynamic time warping (DTW) to produce synthesized machine spoken words (SMSW) at managed complexities and speeds. An ethical approved protocol was conducted with the participation of 25 healthy subjects to create a library of 1500 MBPs corresponding to four different classes. A mean systematic classification accuracy of 91.97 % was obtained using the current configuration. The implications from the study indicate that an improved AAC solution and speaking biometrics decoding could be undertaken in the future.

History

School

  • Mechanical, Electrical and Manufacturing Engineering

Published in

Biomedical Signal Processing and Control

Volume

65

Publisher

Elsevier BV

Version

AM (Accepted Manuscript)

Rights holder

© Elsevier

Publisher statement

This paper was accepted for publication in the journal Biomedical Signal Processing and Control and the definitive published version is available at https://doi.org/10.1016/j.bspc.2020.102303

Acceptance date

24/10/2020

Publication date

2020-11-27

Copyright date

2021

ISSN

1746-8094

Language

en

Depositor

Dr Sijung Hu . Deposit date: 27 November 2020

Article number

102303