This article proposes a virtual reality-based management system for vocal music instruction. Additionally, this article proposes an improved algorithm for automatic vocal main melody extraction. The pitch saliency function calculation method is improved based on the spectral characteristics of vocal music signals in order to reduce the computational complexity and time required to extract the vocal main melody. The model presented in this article has the potential to increase the recognition accuracy of the main melody model, decrease the rate of melody localization false alarms, and increase the overall accuracy of vocal main melody extraction. Additionally, this article incorporates the extraction algorithm into the management system, making it convenient for teachers to use during instruction.