基于FDA的基频建模

Modeling pitch contour based on functional data analysis

提出了一种利用FDA方法对语音基频包络建模的新方法.用B-样条函数对4种声调的单字基频抽取其基频样点进行数据平滑处理,得到平滑后的基频曲线,将平滑后的基频曲线进行2次时间校准处理,得到拟合后的基频曲线.将原始基频与拟合后的基频曲线进行对比,实验结果表明,文中提出的方法建立的基频模型的均方误差为6.47Hz,可应用于语音合成等语音信息处理中.

A novel method for modeling pitch contour with FDA method is presented. By smoothing the pitch＇s samples of four kinds of Mandarin monotone with B-spline basis function, the fitted pitch contour is obtained. Comparing the pitch contours of before alignment and after alignment, the experimental results demonstrated that proposed method can accurately model the pitch contours with 6.47 Hz of mean root error. Proposed method can be applied to speech synthesis.