摘要
由快速傅里叶变换(FFT)初步估计出的语音信号的各阶共振峰频率确定相应带通滤波器的参数,并用该参数对语音信号作滤波处理,对滤波后的信号进行经验模态分解(EMD)得到一族固有模态函数(IMF),按能量最大原则确定出含有共振峰频率的IMF,计算出该IMF的瞬时频率和Hilbert谱即得到语音信号的共振峰频率参数.实验结果表明,与传统方法相比,该方法无须对语音信号进行分帧截断,提高了语音信号共振峰频率估计的时频分辨率和准确性,能够更精确地反映共振峰频率随时间的快速变化.
After being filtered with the band-pass filters with the centre-frequencies obtained by using the fast Fourier transform （FFT） analysis, speech data were decomposea into a set of intrinsic mode function （IMF） using empirical mode decomposition （EMD）. The IMFs containing formant frequencies were then identified according to the energy maximum criteria, and their instantaneous frequencies and Hilbert spectra were calculated, and finally, the formant frequencies of speech data were efficiently determined. The results show that, compared with the conventional formant estimation methods, the method based on HHT not only can provide more clear descriptions of the non-linear and non-stationary characteristics of speech signals, but also gives the speech formant frequencies and their variations with high time-frequency resolution and veracity.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2006年第11期1926-1930,共5页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(60275004)
