The methodology of developing a Verbal Therapeutic Desktop application for ASD Children

The methodology that we suppose to use throughout this project will be discussed in this article.
When we think about the therapies that can be done for autistic children, Verbal therapies get the highest place. Since they are lack with a developed vocabulary and shows less communication skills in their early childhood, that ability should be improved with the involvement of their parents and therapists.

Here we are suggesting a system that could help ASD children to improve their verbal skills by using an interactive way with the use of technology. As the first step to accomplish the above objective we are going to collect a data set of voice recordings of the autistic children. These data set could be consist of different verbal forms that they trying to express their basic needs. And also a data set of voice recordings of people in order to train the system. We will discuss about that later in this article.
So the collected data set is transformed into a machine understandable way by digitizing each and every signal in order to transform all analog signals into its digital form.

Those digitized signals are then subjected to pre-processing to remove the background noises and to get a sharp audio signal for further processing. For that purpose the pre-processing techniques like different filtering methods are used. By using the pre-processed audio signal the related spectrogram is then obtained for further analysis of the digital signal to find the features and to identify the characteristics of the signal.

The next phase is the speech recognition. The recognition is going to be done using an AI model which is designed to be trained by using different audio signals that are in the data set we collected at the beginning of the project. Here we are going to use deep learning methodologies in order to generate an algorithm that is capable of recognizing the voice signals given by the user. Finally an output will be generated using the probabilistic matching of audio signals by subjecting them to the well trained AI model.