中文摘要：

通过分析语音相位的听觉特性，把这种听觉特性运用到基于正弦模型语音编码合成元音的相位处理中，只通过幅频信息和少量的原始相位信息来合成重建语音信号，仿真实验结果表明这种方法能改善合成信号中帧与帧之间的不连续性，使帧与帧之间能平滑过渡，能增加语音自然度，提高语音的合成质量。

英文摘要：

The phase characteristics of human hearing and apply it to the sinusoidal representation of vowel by processing the phases are analyzed. This method uses magnitude-frequency information and only a few phase information to reproduce the speech. Experimental results show that it can improve the disconnection between the frames, make the synthetic speech more natural, and improve the synthetic quality as well.