Bidirectional Encoder Representation from Transformers and Particle Swarm Optimization based Evaluation of Oral Function in Preschool Teachers

Preschool teachers play a vital role in shaping young students’ basic skills and knowledge. Oral communication has aspects like clarity, tone, voice, etc. If early identification of the oral function doesn’t happen, young children might face challenges in communication like pronunciation, clarity, and tone. To overcome these issues this paper and develop a model for automatic evaluation system for oral function of preschool teachers proposed a Bidirectional Encoder Representation from Transformers (BERT) with Particle Swarm Optimization (PSO) for feature selection, efficiently achieving better accuracy. Initially, the process begins with data collection using classroom vocal interaction and engagement in an inclusive preschool named cohort that contains data in audio format. Then, the audio data is preprocessed using noise reduction, and further by incorporating the Natural Language Preprocessing (NLP) technique namely Hidden Markov Model (HMM) the audio data is converted into text data. Then, the converted text data performs feature selection using a Bidirectional Encoder Representation from Transformers (BERT) which selects the relevant features like pitch, frequency, voice, tone, etc. from the converted text. Later, Particle Swarm Optimization (PSO) is used as an optimization technique that optimizes the selected features and gives optimal solutions. The proposed model acquired greater results with an accuracy of 90%.