Enhancing Sanskrit Isolated Word Recognition: A Comparative Analysis of MFCC and SVM Feature Integration
Ashwini Shivaji Ganakwar, Santosh K. Maher, Ratnadeep R. Deshmukh · 2023
This paper presents an innovative approach to the recognition of isolated Sanskrit words, accompanied by the establishment of a comprehensive standardized database system. The core objective of this research is to augment the accuracy of word recognition through the incorporation of cutting-edge techniques. To realize this objective, an elaborate Sanskrit speech corpus was meticulously curated within the confines of our Speech Augmented Human-Computer Interface Lab (HCI). This corpus encompasses isolated Sanskrit digits (ranging from 0 to 9) as well as 15 distinct vowels. The meticulous process employed in this study encompassed the recording of each word on three distinct instances, featuring the active participation of 20 diverse speakers. This selection of speakers was made to encapsulate variations in gender and age within the 20 to 30 age range. By means of methodical experimentation, this research delves into the comparative evaluation and benchmarking of recognition accuracy, utilizing the Mel Frequency Cepstral Coefficients (MFCC) and Support Vector Machine (SVM) techniques. The ensuing insights shed illuminating light on the efficacy of these methodologies in precisely recognizing and discerning Sanskrit words within the formulated database. This endeavor not only constitutes a noteworthy contribution to the realm of Sanskrit speech recognition but also lays the groundwork for prospective advancements in this domain.