基于高斯混合模型的音乐情绪四分类研究

针对音乐情感复杂难以归类的问题,提出了一种在四分类坐标下建立高斯混合模型进行音乐信号归类的研究方法。在建立模型的基础上,创新地为表示情绪特性的轴两端建立模型使其转换成二层分类器进行加权判别。结果表明,为表示情绪特性的轴建立模型且权值分配在0.7和0.3的条件下,音乐的分类工作可以取得最优结果,其结果明显优于直接为每类情绪建立模型的结果。

Music Emotion Four Classification Research Based on Gaussian Mixture Model

For the problem of music emotional complexity and difficult to categorize, we proposed a method to estab-lish Gaussian mixture models in four classifications. On the basis of establish models, we innovated established GMM for shaft at both ends of the emotional model and converted it into two-layer weighted classifier discrimination. The re-sults shows that the GMM for shaft models and weight distribution under the condition of 0.7 and 0.3, the musical work can obtain the best classification result, and the result is better than the result of directly establish models for each type of emotion.