Towards Automatic Detection of Amyotrophic Lateral Sclerosis from Speech Acoustic and Articulatory Samples

Amyotrophic lateral sclerosis (ALS) is a rapid neurodegenerative disease that affects the speech motor functions of patients, thus causes dysarthria. There is no definite marker for the diagnosis of ALS. Currently, the diagnosis of ALS is primarily based on clinical observations of upper and lower motor neuron damage in the absence of other causes, which is timeconsuming, of high cost, and often delayed. Timely diagnosis and assessment for ALS are crucial. Automatic detection of ALS from speech samples would advance the diagnosis of ALS. In this paper, we investigated the automatic detection of ALS from short, pre-symptom speech acoustic and articulatory samples using machine learning approaches (support vector machine and deep neural network). A data set of more than 2,500 speech samples collected from eleven patients with ALS and eleven healthy speakers was used. Leave-subjects-out cross validation experimental results indicate the feasibility of the automatic detection of ALS from speech samples. Adding articulatory motion information (from tongue and lips) further improved the detection performance.