Abstract: According to the non-stationary and nonlinear time-varying characteristics of speech signal, a speech pitch detection method based on Hilbert-Huang transform is presented. It is unnecessary to assume that pitch period is stationary with in any segment, it can decompose signal adaptively and possess high time-frequency resolution（not subject to Heisenberg uncertainty principle）. Firstly, it uses the short-time energy to judge voice and unvoice, then the signal is decomposed into a number of intrinsic mode functions, with Hilbert transform, the instantaneous frequency and instantaneous amplitude of each intrinsic mode function are obtained, the components of the intrinsic mode functions are weighted to emphasize the fundamental frequency information according to the characteristics of pitch, lastly, the square of the autocorrelation is used to detect the pitch. Experiments show that compared with the classical methods, the proposed method provides a higher precision and better robustness.

Key words: pitch detection, Hilbert-Huang transformation, empirical mode decomposition

摘要： 根据语音信号非平稳非线性的时变特点，提出了一种基于希尔伯特-黄变换的基音周期检测法。该方法不需要对语音信号进行短时平稳假设，能自适应地对信号进行分解，具有很高的时频分辨率（不受Heisenberg不确定原理的制约）。利用短时能量对语音进行清浊音判断，再经过经验模态分解将信号分解为若干固有模态函数，然后对每个固有模态函数进行希尔伯特变换求其瞬时幅值与瞬时频率，根据基音特点对分解得到的固有模态函数加权求和突出基音周期信息，最后采用自相关平方法进行基音检测。实验表明，该方法较传统的基音检测法精度有所提高，且鲁棒性较好。

关键词: 基音检测, 希尔伯特-黄变换, 经验模态分解