Articulation Analysis in the Speech of Children with Cleft Lip and Palate

Abstract

Hypernasality is a speech deficit that affects children with cleft lip and palate (CLP). It is characterized by the lack of control of the velum, which causes problems when controlling the amount of air passing from the oral to the nasal cavity while speaking. The automatic evaluation of hypernasality could help in the monitoring of speech-language therapies and in the design of better oriented exercises. Several articulation features have been used for the automatic detection of hypernasal speech. This paper evaluates the suitability of classical articulation features for the automatic classification of hypernasal and healthy speech recordings. Two different databases are considered with recordings collected under different acoustic conditions and with different audio settings. Besides the evaluation of the proposed approach upon each database separately, non-parametric statistical tests are performed to evaluate the possibility of merging features from the two databases with the aim of finding more robust systems that could be used in different acoustic conditions. The results indicate that the proposed approach has a high sensitivity, which indicates that it is suitable to detect hypernasal speech samples. We believe that promising results could be obtained with this approach in future experiments where the degree of hypernasality is evaluated.