Journal of Computer Applications ›› 2012, Vol. 32 ›› Issue (04): 1180-1183.DOI: 10.3724/SP.J.1087.2012.01180

• Typical applications • Previous Articles     Next Articles

Super resolution pitch detection based on LPC and AMDF

WANG En-cheng1,SU Teng-fang1,YUAN Kai-guo2,WU Chun-hua2   

  1. 1. School of Information Engineering, North China University of Technology, Beijing 100144, China
    2. School of Computer, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Received:2011-09-13 Revised:2011-11-10 Online:2012-04-20 Published:2012-04-01
  • Contact: SU Teng-fang
  • Supported by:
    the National Natural Science Foundation Project of China

基于线性预测编码与AMDF的高精度基音检测算法

王恩成1,苏腾芳2,袁开国3,伍淳华3   

  1. 1. 北方工业大学 信息工程学院, 北京100144
    2. 北方工业大学
    3. 北京邮电大学 计算机学院, 北京 100876
  • 通讯作者: 苏腾芳
  • 作者简介:王恩成(1976-),男,安徽淮南人,讲师,博士,主要研究方向:信号与信息处理;苏腾芳(1986-),男,湖南娄底人,硕士研究生,主要研究方向:信号处理与信息安全;袁开国(1982-),男,贵州晴隆人,讲师,博士,主要研究方向:信息隐藏与数字水印;伍淳华(1977-),女,湖北黄冈人,讲师,博士,主要研究方向:智能信息处理、信息灾难备份、云存储。
  • 基金资助:
    国家自然科学基金项目;中央高校基本科研业务费专项资金资助项目;北京市教委面上项目

Abstract: According to the mechanism of speech signal, a super resolution pitch detection algorithm, which combined Linear Predictive Coding (LPC) with Average Magnitude Difference Function (AMDF), was proposed. Firstly, residual of LPC was extracted by linear predictive analysis. Then, cumulative mean normalized difference function and difference signal revision were used to make pitch valley sharper. At last, parabolic interpolation and pitch multiple check were taken to select real pitch period. The experimental results indicate that the pitch detection effect of the algorithm is superior to that of the conventional algorithms. The proposed algorithm conquers half frequency errors, and has good accuracy and robustness under the condition of high Signal-to-Noise Ratio (SNR).

Key words: speech signal, pitch period, Linear Predictive Coding (LPC), Average Magnitude Difference Function (AMDF), Auto Correlation Function (ACF)

摘要: 根据语音信号产生原理,结合线性预测编码(LPC)与平均幅度差函数法(AMDF),提出了一种高精度的基音检测算法。该算法首先利用线性预测分析提取残差信号;然后采用累积平均归一化差分函数与差分信号修正,使基音周期的谷值点更加尖锐;最后利用二次函数拟合与基音周期的倍数检查筛选候选值,得到了准确的基音周期。实验结果表明,与传统方法相比, 该算法的基音检测效果有了明显改善,减少了基音检测中的半频错误,在高信噪比下具有良好的准确性和鲁棒性。

关键词: 语音信号, 基音周期, 线性预测编码, 平均幅度差函数, 自相关函数