Children with Speech Disorders Voice Classification: LSTM and BiLSTM Approach Based on MFCC Features

Speech is an essential aspect of communicating with other people. Children with speech disorders have problems communicating with those around them because they are unable to speak normally due to a speech disorder. Their way of communicating is through sign language, which not everyone understands. Therefore, using LSTM and BiLSTM, the researchers developed a speech classification system based on MFCC features for children with speech disorders. Classifying the voices of children with speech disorders is the goal of this research so that others can understand them and improve on the results of the previous study. This research shows that the LSTM and BiLSTM models, trained with epoch values of 100 and 0.01 learning rate values, produce a prediction accuracy of 85.8% for LSTM and 87.3% for BiLSTM. The epoch and learning rate values were chosen after conducting various tuning experiments on these aspects and achieving the highest accuracy. These two models reach an incredibly satisfactory level of accuracy, so they can be implemented in a system architecture to help communicate with children with speech disorders.