Detecting neuromotor disease from speech articulation

Abstract

Speech articulation may be an important resource to study neuromotor activity in relation with neurological diseases, such as Parkinson’s, Alzheimer’s or Amyotrophic Lateral Sclerosis. Through the present work a biomechanical model for the joint structure of the jaw and tongue is introduced. This model allows putting into relation formant trajectories with jaw-tongue kinematics in terms of dorso-ventral and rostro-caudal movements of the structure. The reconstruction of the absolute velocity of the system from inverse filtering estimates of the first two formants allows the introduction of the velocity distribution histograms as descriptors of articulation behavior. A study case using recordings separated in time from a patient suffering from Amyotrophic Lateral Sclerosis, compared to distributions from a normative control is presented. It may be seen that patient’s distributions show larger contents for low ranges of velocity and smaller contents for high velocity ranges compared to the control distribution. Global comparisons of velocity distribution profiles between the control and patient’s records using Information Theory measurements may be used to monitor illness conditions and progress.