基于分数阶倒谱的带噪语音基频跟踪方法

Pitch tracking of noisy speech using cepstrum based on FrFT

分数阶傅里叶变换(FrFT)是一种同时融合了信号在时域和频域的信息的时频分析工具。利用分数阶傅里叶变换在频率随时间变化的信号进行分析的优势,结合语音信号处理中获得语音信号的声道和激励信息的重要手段之一的倒谱分析,以适当的信干比定义作为搜索准则进行自适应搜索合适的变换阶数,使得基于分数阶傅里叶变换的倒谱更能够体现语音信号频率的时变特性。通过对带噪汉语元音的基频跟踪实验,可以看到基于分数阶傅里叶变换倒谱的语音基频提取方法的效果比传统基于傅里叶变换的倒谱更准确。

The Fractional Fourier Transform（FrFT） is a signal time-frequency representation fusing the information of time domain and frequency domain.We take advantage of good performance of FrFT in analyzing nonstationary signals,especially signals with transition frequency following time evolution,combining FrFT with the cepstrum analysis which is one of the important methods of getting speech vocal tract and excitation information.Appropriate Signal Interference Ratio（SIR） is defined as search criteria in adaptive searching for appropriate transform order.The cepstrum based on FrFT appears more accurate in extracting pitch of noisy speech by searching optimal transform order than traditional cepstrum method for pitch extraction.