Optimizing Window Overlapping Size for Unsupervised Voice Activity Detection: A Comprehensive Review and Analysis
Parshotam Pal, Shilpa Sharma · 2024
Voice Activity Detection (VAD) is a critical task in audio signal processing that is required in many different applications to separate speech parts from background noise. An in-depth evaluation of techniques for maximizing window overlapping size for unsupervised VAD systems is provided in the present study. The difficulties in choosing the most suitable and acceptable window configurations are covered in this study, which also includes balancing frequency and temporal resolutions and taking application-specific needs into consideration. This provides an approach that was suggested in the literature study and includes defined constants, audio segmentation, pre-processing methods, Mel filter bank feature extraction, signal smoothing, and thresholding. The combination of the Teager-Kaiser Energy Operator (TKEO) for improved signal analysis is being investigated. The study summarizes research findings from the literature and emphasizes how visualization and iterative experimentation may improve VAD performance and refine the approach. The development of unsupervised VAD systems for practical applications is simplified by an understanding of signal processing methods, difficulties, and possible areas for future research.