Compact Multimodal Model for Age and Gender Estimation

This study shows a simple deep learning model that can guess a person’s age and gender based on their speech and face. User profiling, surveillance, and interactive systems all need to be able to accurately identify demographics. Traditional models that employ only one sort of data, such images or voice, usually perform poorly in noisy or poorly lit environments. The approach that was suggested employs a small convolutional network (MobileNetV2) to look at pictures of faces and a simple neural network (MLP) to deal with audio data that has already been extracted and saved in CSV format. The model combines both forms of input at the beginning and learns to predict both age and gender at the same time through multi-tasking. The model was 73.25% accurate at guessing gender and 98.86% accurate at remembering it on matched datasets. It also guessed age with an average error of 20.50 years. These results reveal that the system could tell the age and gender of people using low-powered gadgets.