An Intelligent and Adaptive Vocal Teaching System Based on Enhanced Human–Computer Interaction and A2MARL

This paper presents an enhanced human–computer interaction (HCI)-based vocal teaching system that integrates several advanced techniques to improve vocal performance assessment and adaptive teaching. Unlike traditional systems that struggle to account for individual vocal abilities, dynamic performance changes, and learner fatigue, the proposed system synergistically combines Convolutional Winsorization Representation for Pitch Estimation (CWREPE), Short-Time Blackman–Harris Fourier Transform (STBHFT), Gaussian Hidden Markov Emission Probability Model (GHMEPM), and Attention Mechanism-based Multi-Agent Reinforcement Learning (A2MARL). CWREPE ensures accurate pitch estimation, while STBHFT enhances format tracking to capture subtle vocal dynamics. GHMEPM captures the temporal evolution of singing, and A2MARL provides personalized feedback by dynamically adjusting vocal exercises based on learner performance and fatigue. This integrated approach results in a highly adaptive and effective vocal training system, achieving a learning efficiency of 97.23%.