Development of an Intelligent Personal Assistant System Based on IoT for People with Disabilities

Approximately 15% of the world's population suffers from different types of disabilities. These people face many challenges when trying to interact with their home appliances. Various solutions are introduced to increase their quality of life, such as controlling their devices remotely through their voices. However, these solutions use command templates that fail to understand the unstructured or semi-structured command. Many authors have recently integrated intelligent personal assistant (IPA) systems, such as Google Assistant, Siri, and Alexa, with control circuits to exploit the advantages of the NLP of these IPAs to control traditional home appliances. However, this solution still struggles with understanding unstructured commands and requires the internet to be available for controlling the devices. This research proposes a new IPA system integrated with IoT, called IRON, for disabled people to use to control customizable devices with a structured and unstructured voice command. The proposed algorithm receives voice orders from the person in a structured or unstructured form and transforms them into text based on the Google Speech-to-Text API. The natural language processing technique splits the commands into tokens to determine the device name and the command type, whether it is a question about device status or a statement. Afterward, the logistic regression classifies the rest of the tokens as positive or negative to turn on or off the device, then sends the command to a Raspberry Pi to control the device. The proposed IRON system is implemented using logistic regression, naive Bayes, and the support vector machine and is trained on a created dataset consisting of 3000 normal, negative, and unstructured commands. The simulation results show that the IRON system can determine 90% of the device's names for all commands. Moreover, the IRON correctly classifies 100% of the commands as positive or negative within approximately 30 s.