基于情感的家庭音乐相册自动生成研究

随着互联网以及社交网络的发展,电子相册逐渐成为应用广泛的基础服务之一,而如何提高相册的用户体验变得尤为重要.本文提出基于情感的家庭音乐相册自动生成研究,旨在解决为用户喜爱的音乐自动推荐与其情感表达相近的相册图片问题.本文从音乐和图像蕴含的情感出发,音乐上选取梅尔频率频谱系数(MFCC)和相关谱感知线性预测(RASTA-PLP)特征,图像上选取比较直观的颜色特征.在算法上使用了核化典型相关分析(KCCA)方法,该算法解决了图像特征与音乐特征之间异构和跨模态的特征融合问题,实现了音乐相册的自动生成.在实验中,客观评测结果表明,采用KCCA方法在查准率方面高于纯CCA方法;在主观评测中KCCA获得69.45%的满意度,与人工推荐的评价结果(78.09%)比较接近,高于随机推荐和CCA方法的满意度.

Automatic Generation of Family Music Album Based on Emotion

With the development of the Internet and social network,electronic album has gradually become one of the basic services,and how to enhance the user experience of music album becomes particularly important.A study is presented on automatic generation of family music album based on emotion,which is designed to solve the problems of the photo album recommendation for a music that you like based on the same emotion.According to the emotions in the music and images,the representative features both for music and iamges are selected,and the Kernel Canonical Correlation Analysis(KCCA) is employed to study the relevance between the music and images in the same emotion.Finally learned basic mapping model is utilized to realize the automatic generation for the music album.In the experiment,the objective evaluation results shows the KCCA method is higher than that of pure CCA method in precision;and in the subjective evaluation,the proposed KCCA method achieves 69.45％ satisfaction,which is close to the results of manually recommended approach (78.09％) and is higher than the results of random recommended approach and pure CCA approach.