Formant Estimation in Children's Speech and its application for a Spanish Speech Therapy Tool

This paper addresses the problem of how to estimate reliable formant frequencies in high-pitched speech (typical in children), and how to normalize these estimations, independent from vocal tract shape or length. The normalized formant frequencies are used to improve the performance of a Computer-Aided Speech Therapy Tool (CASTT) in Spanish. For this purpose, a study was conducted to see what is the relationship between child's height and their vocal tract length, using traditional technologies in speech processing like linear prediction LPC, homomorphic analysis and modeling of the vocal tract. Results of this study show a high correlation between child's height and their vocal tract length. The study is based on speech from 235 healthy children (110 females and 125 males) which contains Spanish vowels utterances, and enables calibration of a CASTT system for children with speech disorders.