基于GMM的歌曲Morphing算法

Singing Voice Morphing Algorithm Based on GMM

针对业余歌手模仿专业歌手唱歌过程中音色不变的问题,提出一种基于高斯混合模型(GMM)的中文歌曲Morphing算法,采用GMM对语音频谱建模,并通过混合业余歌手和专业歌手的语音频谱,实现歌曲的音色转换。结果显示,混合比例因子k=0或1时,ABX测试正确率均为100%,0<k<1时,ABX测试正确率高于62.5%,并且音色转换后的合成歌曲MOS得分在3.29以上,表明在歌唱模仿的过程中专业歌手的音色也可实现按比例的加入。

As the amateur singer＇s tirabre is not converted in the process of voice imitation from amateur singer to the professional singer, this paper presents a singing voice Morphing algorithm based on Gaussian Mixture ModeI（GMM）. The spectrum parameters are extracted from the source and the target singing voices are modeled by GMM respectively, and the singing＇s timbre is converted by mixing their model parameters. Experiments demonstrate that the accuracy of ABX test can be up to 100% in the case of k=0 or 1, and the accuracy still be higher than 62.5% in the case of 0〈k〈1. The average MOS score of synthetic singing voices after timbre conversion is above 3.29. It shows that the professional singer＇s timbre can be added proportionally during the process of singing imitation.