一种实时基音检测算法

A Real-Time Pitch Detection Algorithm

针对目前基音周期检测实时性的要求,提出了一种基于小波变换的语音基音周期实时检测算法.该算法在提取小波系数极大值过程中利用了小波变换极值与信号突变点之间的关系,将小波域波形与时域波形相结合,采取每次搜索以前一个小波系数极值点作为新的基准的自适应基准方式,并利用了平均能量、过零率、历史峰值幅度、当前峰值估计等多特征参数.实验结果表明,该算法在2.5 ms时间内可以准确捕捉并检测到新的基音脉冲位置,而且对语音和残差信号均取得了较好的结果.

A real-time pitch detection algorithm for speech signals is presented. Combining the wavelet waveform with the time-domain waveform, adopting adaptive methods which get a new benchmark by researching for the latest wavelet transform extreme point and making use of multi-characteristic parameters, such as average energy, zero-crossing rate, old peak-amplitude and current peak estimation, the algorithm extracts the wavelet coefficient maxima. Experimental results show the algorithm can detect the pitch pulse in 2.5ms and is suitable for speech and residual signals.