Synergistic Fusion of Deep Learning Techniques for Holistic Analysis of Age Group, Gender and Emotion Prediction from Speech

Voice-based recognition systems have garnered significant attention, owing to their versatile applications in domains such as human-computer interaction, virtual assistants, and affective computing. This research aims to address the challenges associated with concurrent processes and escalating latency in individual models by delving into the intricacies of simultaneous age-group, gender, and emotion recognition from speech data. To surmount these impediments, we present a comprehensive investigation employing three distinct modeling approaches: standalone models for age group, gender, and emotion recognition; sequential models wherein the output of one model serves as input for the next; and an integrated model proficient in concurrent detection of age-group, gender, and emotion. Drawing inspiration from video understanding techniques, the integrated model seeks to streamline recognition processes and minimize latency. Experimental datasets, including the Ryerson Audio-Visual Database of Emotional Speech (RAVDESS), Common voice dataset and Crowd-Sourced Emotional Actors Database (CREMA-D), are lever age-group for rigorous analysis. Evaluation metrics encompassing accuracy, latency, and memory usage-group are employed to compare the performance of the diverse models.