Title: Improving the computational complexity and word recognition rate for dysarthria speech using robust frame selection algorithm

Authors: Garima Vyas; Malay Kishore Dutta; Jiri Prinosil

Addresses: Department of Electronics and Communication Engineering, Amity University, UP, India ' Department of Electronics and Communication Engineering, Amity University, UP, India ' Faculty of Electrical Engineering and Communication, Brno University of Technology, Czech Republic

Abstract: Dysarthria is a speech syndrome caused by the neurological damage in motor speech glands. In this paper, a robust frame selection algorithm has been employed to recognise the dysarthria speech with less time consumption. This algorithm determines the more informative frames which in turn reduce the size of feature matrix used for recognising the speech. This method results in a significant reduction in computational complexity without compromising with the word recognition rate (WRR) which may support a real time application. The amalgamation of four prosodic features: Mel frequency cepstral coefficients (MFCCs), Log of energy per frame, differential MFCCs and double differential MFCCs has been used for training and testing the Hidden Markov Models (HMMs) for speech recognition. Several try-outs were performed on the high, medium and low intelligibility audio clips with a vocabulary size of 29 isolated words. The time complexity of the whole system is reduced up to 54.8% with respect to the time taken by the system without implementing RFS. The proposed scheme is gender, speaker and age independent.

Keywords: dysarthria; hidden Markov models; MFCCs; robust frame selection; speech intelligibility; speech recognition.

DOI: 10.1504/IJSISE.2017.086037

International Journal of Signal and Imaging Systems Engineering, 2017 Vol.10 No.3, pp.136 - 145

Received: 26 May 2016
Accepted: 22 Apr 2017

Published online: 21 Aug 2017 *

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