Automatic Anomaly Detection for Dysarthria across Two Speech Styles: Read vs Spontaneous Speech

Automatic Anomaly Detection for Dysarthria Across Two Speech Styles: Read vs Spontaneous Speech Imed Laaridh1,2, Corinne Fredouille1, Christine Meunier2 1University of Avignon, CERI/LIA, France 2Aix Marseille Universit´e, CNRS, LPL UMR 7309, 13100, Aix-en-Provence, France imed.laaridh@alumni.univ-avignon.fr, corinne.fredouille@univ-avignon.fr, christine.meunier@lpl-aix.fr Abstract Perceptive evaluation of speech disorders is still the standard method in clinical practice for the diagnosing and the following of the condition progression of patients. Such methods include different tasks such as read speech, spontaneous speech, isolated words, sustained vowels, etc. In this context, automatic speech processing tools have proven pertinence in speech quality evaluation and assistive technology-based applications. Though, a very few studies have investigated the use of automatic tools on spontaneous speech. This paper investigates the behavior of an automatic phone-based anomaly detection system when applied on read and spontaneous French dysarthric speech. The behavior of the automatic tool reveals interesting inter-pathology differences across speech styles. Keywords: Dysarthria, automatic speech processing, anomaly detection, read speech, spontaneous speech, speech disorders 1. Introduction Dysarthria is a motor speech disorder resulting from neu- rological damages located either in the central or in the peripheral nervous system. This may lead to disturbances in any of the components involved in the speech produc- tion, including respiratory, phonatory, resonatory, articu- latory and prosodic elements. Consequently, this may be reflected by weakness, spasticity, incoordination, involun- tary movements, or abnormal muscle tone, depending on the location of the neurological damage. Dysarthric speech has been studied according to different axes : perceptual evaluation of speech alterations for dysarthria classification (Darley et al., 1969; Duffy, 2005; Darley et al., 1975), dif- ferent speech tasks (Van Lancker Sidtis et al., 2012; Kem- pler and Van Lancker, 2002), perceptual measurement of dysarthria severity, notably related to the speaker’s intelligi- bility (Enderby, 1983; Yorkston et al., 1996; Hustad, 2008; Lowit and Kent, 2010) or articulatory or/and acoustic anal- ysis (Kent et al., 1999; McAuliffe et al., 2006; Rosen et al., 2006; Green et al., 2013; Whitfield and Goberman, 2014) in order to observe and characterize the effects of dysarthria in the speech signal. These studies aim at helping clinicians in their knowledge of the speech impairment and its clinical evaluation, crucial for following the condition progression of patients in the case of treatment or/and of speech reha- bilitation to enhance them. In this context, automatic speech processing approaches have been seen, very early, as potential solutions to pro- vide objective tools to deal with speech disorders (Ferrier et al., 1992; Rosen and Yampolsky, 2000; Carmichael, 2007; Middag et al., 2009; Nuffelen et al., 2009; Kim and Kim, 2012). In addition, they can also help people with speech disorders in their everyday life through Alternative and Augmented Communication (AAC) tools, involving auto- matic speech recognition for instance (Green et al., 2003; Parker et al., 2006; Rudzicz et al., 2012; Christensen et al., 2013a; Christensen et al., 2013b). In the literature, the set of acoustic-perceptual cues in- cluding imprecision of consonants, distorsion of vowels, slow rate, monopitch, monoloudness, hypernasality is com- monly used to characterize …

Content truncated to 500 words