Preliminary steps to validate Audapter, a software application for online tracking of children’s formants

Cheung, Stephanie 1, 2; Thompson, Kristen 1; Orlandi, Silvia 1; Yunusova, Yana 3, 4, 5; Beal, Deryk 1, 2, 4, 5

1. Holland Bloorview Kids Rehabilitation Hospital; 2. Institute of Biomaterials and Biomedical Engineering, University of Toronto; 3. Sunnybrook Health Sciences Centre; 4. Rehabilitation Sciences Institute, University of Toronto; 5. Department of Speech Language Pathology, University of Toronto

The study of online speech motor planning and speech motor learning depends on the accurate detection and tracking of children’s vowel formant frequencies in near real-time. Formant tracking in children is, however, an extremely challenging task. Audapter is a research software application that has been used to detect and track children’s formants for this purpose, but has never been validated for children’s speech. Our aim was to validate Audapter using natural and synthesized samples of children’s speech. We calculated Audapter’s performance on natural speech samples from the North Texas Vowel Database and on synthetic speech samples generated by our group. Audapter detected the presence of formants in 77% of the from the North Texas Vowel Database. Performance on synthesized speech samples reveals that different vowels are detected with different accuracy, a potential concern for common vowel perturbation study protocols. Further work is needed to improve Audapter's performance and to determine optimal parameters for robust research.