Dépôt numérique

Acoustic and prosodic analysis of pre-verbal vocalizations of 18-month old toddlers with autism spectrum disorder.


Téléchargements par mois depuis la dernière année

Plus de statistiques...

Bedoya Jaramillo, Stefany (2017). Acoustic and prosodic analysis of pre-verbal vocalizations of 18-month old toddlers with autism spectrum disorder. Mémoire. Québec, Université du Québec, Institut national de la recherche scientifique, Maîtrise en télécommunications, 110 p.

[thumbnail of Bedoya Jaramillo, Stefany.pdf]
Télécharger (1MB) | Prévisualisation


Autism Spectrum Disorder (ASD) covers a wide spectrum of symptoms with the main ones relating to problems with social communication and interaction. Definite ASD diagnosis is based on the presence of certain symptoms and their severity levels and, according to current standards, occurs typically at 36 months of age. Recent statistics show that about 1 in 68 children are diagnosed with autism and there is a recurrence rate of 18.7% for the biological siblings of autistic individuals. As such, early detection is critical, as it may allow for intense therapy to be initiated, thus tapping into a young brain’s plasticity properties and increasing odds of success. Today, researchers and clinicians have joined efforts to understand and identify new markers of the disorders, thus allowing for early diagnosis, ideally around 18 months of age. To this end, acoustic analysis of toddler vocalizations has emerged as a promising area, even for pre-verbal children. Prosodic and acoustic disorders have been reported for babble and speech-like vocalizations. As such, pitch, energy and voice quality related features have been explored for early ASD diagnosis. In this work, we build upon these findings and propose the use of wavelet-based and speech modulation spectral features for ASD diagnosis based not only on speech-like verbalizations, but also on cries, laughs, and other sounds made by the toddlers. We show that the proposed features are complementary to existing ones and, on a cohort of forty-three 18-month old toddlers, a support vector machine classifier was capable of correctly discriminating the ASD group from the typically-developing toddlers with accuracies above 80%, thus outperforming existing methods. More importantly, we show that with these new features, vocalizations such as cries, squeals, whines and shouts showed to be more discriminative than babble and speech-like vocalizations. It is hoped that these findings will lead to more accurate early diagnosis of ASD symptoms.

Type de document: Thèse Mémoire
Directeur de mémoire/thèse: Falk, Tiago H.
Co-directeurs de mémoire/thèse: O’Shaughnessy, Douglas
Mots-clés libres: autism spectrum disorder; diagnosis; prosody; wavelets; speech modulation spectrum
Centre: Centre Énergie Matériaux Télécommunications
Date de dépôt: 29 janv. 2018 21:51
Dernière modification: 29 janv. 2018 21:51
URI: https://espace.inrs.ca/id/eprint/6655

Gestion Actions (Identification requise)

Modifier la notice Modifier la notice