Abstract: A short time pitch detection algorithm based on waveform extremum is proposed. Speech extremum is extracted after denoised and voiced / unvoiced detected, then the waveform extremum is modified. Magnitude difference is calculated, and pitch is detected by a certain pitch criterion. The algorithm, with high accuracy, does not need divided window processing, and main calculation is waveform extremum processing and has a smaller amount of calculation than common AMDF. Experimental results show that this method can get accurate pitch and is suitable for real-time processing.

Key words: Waveform Extremum Method（WEM）, pitch period, Average Magnitude Difference Function（AMDF）, Length Variable-Average Magnitude Difference Function(LV-AMDF), Circular Average Magnitude Difference Function（CAMDF）

摘要： 提出了一种基于波形极值的短时基音检测算法。在对语音数据进行预处理、清浊音判断后，提取语音波形极值，进行波形修正，计算幅度差，采用一定的基音判断准则查找基音。该算法无需对音频数据做分窗处理，且主要运算是对波形极值进行处理，比常用的AMDF有着更小的计算量，且有较高的精度。实验结果表明，利用该方法能够很好地得到基音周期，适宜进行实时处理。

关键词: 波形修正算法（WEM）, 基音周期（PITCH）, 平均幅度差函数（AMDF）, 变长平均幅度差函数（LV-AMDF）, 循环平均幅度差函数（CAMDF）