Acoustic to kinematic projection in Parkinson’s disease dysarthria
Introduction
The influence of neurological and cognitive processes on speech is a well-established and recognized fact [[1], [2], [3]]. Many studies in the last decade have explored diverse signals such as EEG, MEG, fMRI and other non-invasive methods to provide new insights into the vocal production process [4,5]. This is of particular interest when investigating neurological diseases (cognitive and neuromotor) such as Alzheimer, PD, Amyotrophic Lateral Sclerosis (ALS), Huntington’s Chorea, and others related [6]. Speech allows the contactless remote recording on smart terminals, as phones, tablets or laptop computers. Speech offers the added value of mapping acoustic estimates to neuromuscular activity, providing an advantage in the detection and monitoring of diseases dependent on neuromotor pathway transmission remotely [7]. A comprehensive study on the effects of PD on speech [8,9] could provide insights into the underlying physiology, associating speech characteristics to the physical manifestations of the disease. This can be achieved through the study of phonation, articulation, prosody and fluency [10] which would offer valuable information on the activity of specific brain areas involved in speech production, such as motor planning, premotor and motor, and working memory. There is an unmet need to establish a robust and reliable methodology to map estimates extracted from the speech acoustics to motor actions in certain muscles involved in speech articulation and production. One such example is the masseter muscle, responsible of raising the lower jaw. Such a projection methodology is proposed in this research work to transform speech formant dynamics to articulatory kinematics [11,12]. First proposals of an inverse model (relating formant dynamics and articulation) were presented; as a result several indicators were developed to encompass articulatory movements from speech alone (e.g. Absolute Kinematic Velocity, AKV) [13,14]. The problem with these first attempts was the lack of a robust model parameter estimation. This led to further exploratory work, were the relationships between sEMG, accelerometry and Speech were investigated [15]. After an in-depth study of the affectations of PD on these biometric signals [16], the conclusions were applied to the characterization of PD hypokinetic dysarthria [17,18].
The aim of the present study is to provide an insight into acoustic-to-kinematic projection, which could eventually allow to extract and transfer acoustically relevant articulation features to neuromotor actions, to be used in the characterization and monitoring neuromotor activity in specific diseases such as PD which is used as testbed in this study, this being the objective of future research already in progress. This approach can provide new insights into the physiological voice production mechanism and tentatively assign any effects of PD on specific vocal production model components. Such a model would add new semantic value over other standard approaches in the state of the art. In this regard, the previously proposed mapping model is reformulated in terms of time and space variables to allow a dynamic description of the model coefficients to be used in further mapping processes in remotely monitoring PD. This description includes also estimations from HCs.
The paper is structured as follows. In Section 2 the acoustic-mechanical model proposed is reformulated in the time-space domain to allow a more robust estimation of its coefficients. Section 3 is devoted to describe the data acquisition framework, platform and protocols, the biometrical data of participants, and the estimation methodology for the model coefficients in the space-time domain. The results derived from the PD and HC subsets are presented and described in Section 4. Section 5 is devoted to discuss the robustness of the methodology considering the results, and to analyze their impact and limitations in possible online applications. The study’s key findings are summarized in Section 6.
Section snippets
The neuromechanical model of the lower jaw articulation
The present study is based on a simplified jaw-tongue articulation model [16] which is known to be representative of PD dysarthria [18]. It allows to create a relationship between acoustic and kinematic variables relating the first two formants F={F1, F2} to the horizontal and vertical coordinates S={xr, yr} of the joint Jaw-Tongue Reference Point (PrJT) in the sagittal plane This point represents the center of moments of the biomechanical system integrated by the maxillary bone, tongue and
Data acquisition framework
The study cohort comprises 8 PD participants (four male, four female, all Spanish native speakers, stage 2 in H&Y scale) who were recruited from a PD patient association in the metropolitan area of Madrid (Asociación de Pacientes de Parkinson de Alcorcón y Móstoles, APARKAM). For comparison purposes four male and four female healthy control participants have been included in the study. The biometrical data of participants are given in Table 1.
The study was approved by the Ethical Committee of
Data recording examples
The speech signal, the sEMG and the 3 acceleration channels from two repetitions of the […aja…] by a female HC participant (CF1) are shown in Fig. 3 as an illustrative example. The sEMG signal has been included in the plots (channel b) with the purpose of witnessing that the acceleration and speech signals are concordant with the action of the masseter.
Similarly, the same set of recordings from one of the PD female participants included in the study (PF1) are shown in Fig. 4.
The repetition of [
Discussion
In Section 3 an inverse linear model based on an acoustic to kinematic projection has been presented. This model has been validated by the results shown in Section 4, consequently the following findings may be highlighted:
- •
The relationship between acoustic to kinematic variables (ΔF to ΔS) has been established and may be explained using the inverse model described in expression (1).
- •
The initial estimation of the model weights has been carried out using least squares linear regression.
- •
A
Conclusions
The present study has been conceived to provide further insights into the acoustic-to-kinematic model of the jaw-tongue articulation joint, based on preliminary approaches. In summary, the key findings derived from this study are:
- •
An acoustic-to-kinematic model to predict the jaw-tongue joint kinematics from acoustic dynamics expressed in formants has been examined in depth, with special emphasis on weight estimation procedures.
- •
A weight estimation refinement method based on an iterative gradient
CRediT authorship contribution statement
A. Gómez: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Visualization. A. Tsanas: Validation, Formal analysis, Resources, Writing - review & editing, Supervision, Project administration, Funding acquisition. P. Gómez: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing - review & editing, Visualization, Supervision. D. Palacios-Alonso: . V. Rodellar:
Acknowledgments
This research has been funded by grants TEC2016-77791-C4-4-R (MINECO, Spain) and CENIE_TECA-PARK_55_02 INTERREG V-A Spain-Portugal (POCTEP). The authors thank Asociación de Parkinson de Alcorcón y Móstoles (APARKAM), and Azucena Balandín (director) and Zoraida Romero (speech therapist) for their help and advice.
Declaration of Competing Interest
The authors report no declarations of interest.
References (27)
- et al.
Vowel articulation in Parkinson’s disease
J. Voice
(2011) - et al.
Robust and complex approach of patohogical speech signal analysis
Neurocomputing
(2015) - et al.
Assessing correlations between lingual movements and formants
Speech Commun.
(2013) - et al.
Articulatory-acoustic vowel space: application to clear speech in individuals with Parkinson’s disease
J. Commun. Disord.
(2014) - et al.
Characterization of Parkinson’s disease dysarthria in terms of speech articulation kinematics
Biomed. Signal Proc. and Control
(2019) - et al.
Progession of voice and speech impairment in the course of Parkinson’s disease: a longitudinal study
Parkinsons Dis.
(2013) Multiple factors are involved in the dysarthria associated with Parkinson’s disease: a review with implications for clinical practice and research
J. Speech Lang. Hear. Res.
(2014)- et al.
Imprecise vowel articulation as a potential early marker of Parkinson’s disease: effect of speaking task
J. Acoust. Soc. Am.
(2013) - et al.
A deep learning approach for Parkinson’s disease diagnosis from EEG signals
Neural Comput. Appl.
(2018) Use of Magnetoencephalography (MEG) to study functional brain networks in neurodegenerative disorders
J. Neu. Sciences
(2010)