基于波内频率调制的音色模型研究以及在单通道音源分离中的应用

非线性欠定音源分离技术是音频技术中的热点和难点问题之一。本课题针对音色的非线性本质，从非线性信号的波内频率调制现象入手，对乐器声音信号的频率结构进行分析，通过Hilbert-Huang Transform得到对信号的非线性表征，并以此为基础提出新的音色模型。我们的初步研究结果表明，这种新的音色模型可以将音源的非线性部分转化为线性部分，从而将大大提高非线性欠定音源分离的效果和鲁棒性。本课题的研究对于音乐信息理论，核心软件产品和新一代音频编码技术等方面具有重要的意义。

The under-determined blind source separation is one of the most important problem of audio signal processing technology. In our work, we use a time-frequency joint analysis method named Hilbert Huang Transform (HHT) to review the instruments sound signal structure and find out a novel timbre expression based on the intra-wave modulation. Furthermore, a novel timbre model based on Intra-wave Frequency Modulation is developed. The tentative experiments show that the model can transfer the nonlinear part of the signal into linear expression, which will improve the results and performance of the under-determined blind source separation. Our work has the potential to benefit music information theory, core software production and next generation audio code technology.