Abstract

This paper presents a novel falsetto detection algorithm designed to enhance the performance of Voice Gender Recognition (VGR). By incorporating Signal Processing techniques with insights from vocal pedagogy, our algorithm identifies falsetto in singing voice data to reduce gender identity ambiguity in vocal analysis. We used a pre-trained Deep Learning VGR model to assess the effectiveness of our algorithm. Experiments with various parameter settings demonstrate that the proposed algorithm reduced false positives in male voice detection and improved the VGR F1 score by a maximum of 5.3% for male voices and 2.6% for female voices. Our findings also highlight potential advancements in falsetto detection and provide insight for improving applications such as Voice Age Detection.