Intelligent Extraction And Application Of Musical Rhythm In Chinese Pop Songs

The extraction of musical rhythm,technically known as beat recognition,is a core technology in music signal processing.Beat recognition plays a crucial role in multimodal applications.It provides an accurate temporal reference for music information retrieval and music generation,and is instrumental in fields such as music accompaniment,dance choreography,and music education.Accurate beat recognition enhances the expressiveness and interactivity of musical works,offering substantial support for music creation and performance.Additionally,its potential applications extend to music therapy and sports training.This thesis aims to bring new technological breakthroughs to relevant fields by improving the accuracy and real-time performance of beat recognition,thereby promoting the integration and development of music technology with other domains.In this thesis,a comprehensive analysis and assessment of the current state-of-the-art beat recognition methods are conducted.Several measures are proposed to enhance the performance of beat recognition tasks:(1)Addressing the issue of insufficient training data in beat recognition,a noisy teacher-student training framework is introduced,combining semisupervised and supervised learning to enhance model generalization;(2)To handle musical beat features across different time scales,the network backbone is improved by using multiple dilation rates in dilated convolutional layers,enhancing model performance;(3)For the needs of music with variable time signatures,a new post-processing method is developed to support variable time signature downbeat recognition,expanding the application scope of beat recognition tasks.To validate the improvements,ablation studies are conducted to verify the effectiveness of each measure,demonstrating performance comparable to the most advanced methods,with broader application scenarios.Additionally,a new Chinese pop song beat annotation dataset is constructed and presented.Finally,this thesis introduces an AI music beat recognition platform based on the proposed algorithm as an application of the algorithm.