Abstract: Traditional speech spectral smoothing algorithm is often based on source-filter model，it needs analyze the speech signal into an excitation signal and a vocal filter.This procedure will cause error，and degrade the quality of the synthesized speech.A speech spectral smoothing algorithm based on Fourier transform is studied in this paper.It can avoid errors caused by the procedure of analyzing.Experiment result shows that this method can smooth the spectra of speech well and get high quality synthesized speech，meanwhile.

Key words: concatenative speech synthesis, spectral smoothing, spectral modification

摘要： 传统的基于源-滤波器模型的语音频谱平滑算法，需要将语音信号分解为激励源和对应的声道滤波器，这样做会引入误差，最终导致合成语音质量的下降。研究了一种基于傅立叶变换的语音频谱平滑算法，避免了上述的分解步骤。实验表明，这种方法能够较好地进行语音频谱平滑并且使合成语音的质量没有受到太大影响。

关键词: 拼接语音合成, 频谱平滑, 频谱修正