Results from Using Automatic Speech Recognition in Cleft Speech Therapy with Children

Abstract

Most children with cleft are required to undertake speech therapy after undergoing surgery to repair their craniofacial defect. However, the untrained ear of a parent can lead to incorrect practice resulting in the development of compensatory structures. Even worse, the boring nature of the cleft speech therapy often causes children to abandon home exercises and therapy altogether. We have developed a simple recognition system capable of detecting impairments on the phoneme level with high accuracy. We embed this into a game environment and provide it to a cleft palate specialist team for pilot testing with children 2 to 5 years of age being evaluated for speech therapy. The system consistently detected cleft speech in high-pressure consonants in 3 out of our 5 sentences. Doctors agreed that this would improve the quality of therapy outside of the office. Children enjoyed the game overall, but grew bored due to the delays of phrase-based speech recognition.