论文部分内容阅读
和弦的感知是音乐自动标注的基础,对于音乐结构分析及旋律分析等任务有重要的作用,因此成为音乐信息检索(M IR)领域的热点之一。本文根据音乐认知心理学原理,提出一种基于人工神经网络(ANN)的和弦实时感知方法:首先利用常数Q变换(CQT)对音乐信号进行时频变换,并在所得到的谱上实施音符起始点检测、音高校准及基频分析等技术以增强系统的鲁棒性,之后定义了一种全新的音级分布矩阵(PCDM)特征,最后利用ANN模拟人脑认知过程并通过半监督学习方法对和弦进行感知。在多种风格音乐上进行的初步实验表明,所提出的方法以可以接受的计算时间消耗取得接近60%的识别率,与本领域先进方法的效果不相上下。
Chord perception is the basis of automatic music annotation, and plays an important role in the task of music structure analysis and melody analysis. Therefore, it has become one of the hot spots in the field of music information retrieval (MIR). Based on the theory of music cognitive psychology, this paper proposes a chord real-time sensing method based on Artificial Neural Network (ANN). Firstly, constant frequency Q transform (CQT) is used to transform the music signal in time and frequency and perform musical notes on the resulting spectrum Starting point detection, pitch calibration and fundamental frequency analysis to enhance the robustness of the system. Then, a new PCD is defined. Finally, ANN is used to simulate the process of human cognition and through semi-supervised Learning method to perceive the chord. Preliminary experiments on multi-genre music show that the proposed method achieves an approaching 60% recognition rate with acceptable computational time consumption, comparable to the state-of-the-art methods.