Abstract: Pitch period is an important parameter of speech signal，and it has been applied in many domains.A pitch detection method based on spectral subtraction is proposed，noised speech is denoised by spectral subtraction method，and then the autocorrelation function of the LPC predictive residual and the cepstrum of the autocorrelation function are calculated.The result of the simulation shows that using this improved method in pitch detection，the effect can have a distinct improvement compared to the classical pitch extraction method.

Key words: pitch detection, spectral subtraction, cepstrum, autocorrelation function

摘要： 基音周期是语音信号的一个重要参数，它在多个领域有着广泛的应用。提出了一种基于谱减法的基音检测算法：先用谱减法对带噪语音去噪，然后再求语音LPC预测残差的自相关函数及自相关函数的倒谱。仿真结果表明，利用这种改进算法做基音周期检测，检测效果会比传统倒谱检测方法有明显改善。

关键词: 基音检测, 谱减法, 倒谱, 自相关函数