In the United States, speech, voice, and language disorders affect approximately 1 in 14 children aged 3-17 years old. If impairments are left undiagnosed, they could impact communication, academics, social interactions, and overall quality of life as the child develops. There is a need for objective, quantifiable diagnostic methods for children that can offer guidance while informing personalized and effective speech training strategies.

Our overarching goals are to identify characteristic markers of voice in children with impaired speech by investigating extracted acoustic features and to distinguish between different speech disorders using trained machine learning (ML) models. We will achieve our research goals by characterizing the key features of select pediatric speech disorders (Aim 1), distinguishing the key features of each using ML models (Aim 2), and exploring methodologies used for speech screening and training in children (Aim 3). The features we extract will support the characterization of speech patterns in pediatric patients. The extracted features will subsequently be used in conjunction with classification models including Support Vector Machines (SVM), Random Forest (RF), Convolution Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM). Overall, this project aims to provide data-driven metrics to enhance diagnostic approaches toward disorders involving pediatric speech.