Vocalization Recognition of People with Profound Intellectual and Multiple Disabilities (PIMD) Using Machine Learning Algorithms

We investigate vocalization recognition for people with Profound Intellectual and Multiple Disabilities using various machine learning algorithms. The amount of training data available for people with PIMD is typically significantly limited. Due to this fact, data augmentation process was used. Various types of Machine Learning algorithms were tested: k-NN, NB, DT, RDF, MLP and LSTM. During research we also tested various regularization techniques to improve recognition performance. The best results were obtained in case of MLP network with dropout and batch normalization: 90%.