一种高精度的基音提取方案  

A High Precision Solution of Pitch Detection

作　　者：黄海亮[1] 谢康林 杜平 吴边 

机构地区：[1]上海交通大学软件学院,上海200030 [2]上海交通大学计算机科学与工程系,上海200030 [3]英特尔中国软件中心,上海200336

出　　处：《计算机工程》2004年第B12期343-345,372,共4页Computer Engineering

摘　　要：为了得到准确的基音曲线,该文提出了一个有效的解决方案,通过利用人工神经网络进行浊音判决,以及计算基音频率时利用基于 自相关函数的动态规划算法,有效地克服了在语音信号清音和无声段错误的基音提取以及在浊音段的2倍频或1／2倍频错判。实验证明,利用 该文提出的方案,可以大大提高基音频率提取的准确度,从而得到非常准确的语音基频曲线。To get a precise pitch contrail, an efficient solution is presented in this paper. The solution introduces a voiced/unvoiced decision step based on artificial neural networks and then applies the dynamic programming algorithm based on autocorrelation function method to extract the pitch value. It avoids the wrong pitch detection in unvoiced/silence segment of speech signal and wrong pitch value judgment in the point of double or half pitch value. The performance of this system shows that this solution greatly improves the precision of pitch detection.

关 键 词：基音频率 自相关函数 人工神经网络 动态规划 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]