Automated Diagnosis and Assessment of Dysarthric Speech Using Relevant Prosodic Features  

Oleh:

Kadi, Kamil Lahcene ; Selouani, Sid Ahmed ; Boudraa, Bachir ; Boudraa, Malika

Jenis: Article from Books - E-Book
Dalam koleksi: Transactions on Engineering Technologies: Special Volume of the World Congress on Engineering 2013, page 529-542.
Topik: Dysarthria; GMM; LDA; Nemours database; Prosodic features; Severity-level assessment; SVM
Fulltext: 38_978-94-017-8831-1_Kadi_Selouani_Boudraa.pdf (560.39KB)

Isi artikel

In this paper, linear discriminant analysis (LDA) is combined with two automatic classification approaches, the Gaussian mixture model (GMM) and support vector machine (SVM), to automatically assess dysarthric speech. The front-end processing uses a set of prosodic features selected by LDA on the basis of their discriminative ability, with Wilks’ lambda as the significant measure to show the discriminant power. More than eight hundred sentences produced by nine American dysarthric speakers of the Nemours database are used throughout the experiments. Results show a best classification rate of 93 % with the LDA/SVM system achieved over four severity levels of dysarthria, ranged from not affected to the more seriously ill. This tool can aid speech therapist and other clinicians to diagnose, assess, and monitor dysarthria. Furthermore, it may reduce some of the costs associated with subjective tests.

Opini Anda

Klik untuk menuliskan opini Anda tentang koleksi ini!

Kembali