基于Hilbert-Huang变换的基音周期提取方法

Pitch detection method based on Hilbert-Huang Transform for speech signals

提出了一种基于Hilbert-Huang变换的语音信号基音周期提取方法。该方法无须对语音信号进行分帧截断。语音信号直接进行Hilbert-Huang变换后,通过基音频率搜索处理得到基音频率及其随时间的变化。实验结果表明,与传统的基音周期提取方法相比,该方法既能真实描述语音信号的非平稳非线性特性,又能提高基音周期提取的准确性和分辨率。

A speech pitch detection method based on Hilbert-Huang Transform （HHT） is presented in this paper. With the speech processing method, truncating and separating of the speech data into frames are no longer needed. Applying the Hilbert-Huang Transform and further post-processing on the speech signals, the speech fundamental frequency and its variation with respect to time are obtained. The analysis results presented in this paper show that, compared with conventional methods of the pitch detection, the method based on HHT not only gives the direct and true description and representation of the non-linear and non-stationary characteristics of speech signals, but also provides a higher accuracy and resolution in speech pitch detection.