摘要
声门激励信号是语音信号的源信号,可用于语音特征参数的有效提取。研究了从观测语音获取声门激励的两种方法——线性预测法和倒谱法;用实际录制的语音做计算机仿真实验,比较了两种方法的性能和特点。结果表明倒谱法获取声门激励、由它提取基因周期等激励特征参数的精度高,但计算量相对较大;线性预测法由于采用高效算法,不仅获取声门激励的速度快,而且可同时获取声道模型参数、语音功率谱等重要参数,是获取声门激励的常用方法。
glottal pulses is source of speech signal.It is used in pitch detection.This paper elaborated two method of obtained glottal pulses,such as LPC and cepstrum.The results of the simulated experiment on the algorithm for both show that cepstrum method is high precision and large computation.Further it indicated LPC method is common and efficient algorithm in obtained glottal pulses and parameters of vocal tract's model and power spectrum.
出处
《电脑开发与应用》
2010年第8期13-15,共3页
Computer Development & Applications