Water chicken swarm optimization-based deep segmental neural network for spoken term detection using bayesian filtering

Abstract

The spoken language processing system becomes a challenging task in the mix-lingual and multilingual scenario. The typical features, like nasalized vowels, aspirated plosives and unvoiced retroflex plosives from the spoken languages faces complex issues in spoken term detection due to the lack of knowledge about training data and language of interest. Hence, an effective and optimal spoken term detection mechanism named Water Chicken Swarm Optimization-based Deep Segmental Neural network (WCSO-based DSNN) is proposed in this research to detect the spoken words from speech signals. The proposed approach is evaluated on the challenging multilingual Quesst2014 dataset. The proposed WCSO algorithm is designed by integrating the Water Wave Optimization (WWO) algorithm, with the Chicken Swarm Optimization (CSO) algorithm. The DSNN classifier is facilitated to detect the spoken words from the speech signal based on the fitness function. The distance between the query speech signal and the segmented spoken words is measured using the Cosine distance. Moreover, the proposed method obtained better performance for minCnxe with the value 0.5531 for type 3 evaluation queries for multilingual languages. The experimental results demonstrate that the proposed method performs better compared to several state-of-the-art Speech to Speech matching methods. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.