摘要
为了合成可变音长的语音，若简单地将基音周期拉长或缩短，则会造成基频的降低或升高，从而造成音调的失真．本研究对原始语音的基音变化趋势进行了拉格朗日内插和最小二乘法拟合，并保存其内插及拟合的系数；在合成时，利用系数来恢复并插入基音从而改变发音长度，避免了音调的失真．
On purpose to synthesize the variable long speech, only prolonging or shortening the tonic period will lead to a drop or rise in the base frequency, resulting in the distorted tone. This paper deals with the syntheses of variable long speeches in virture of Lagrangian interpolation and least square fitting by preserving their coefficients, inserting the tonic and changing the length, thereby avoiding the tone distortion.
出处
《华南理工大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
1998年第3期27-31,共5页
Journal of South China University of Technology(Natural Science Edition)
关键词
线性预测编码
语音合成
内插
拟合
linear predictive coding
speech synthesis
interpolation
fitting
