刘宇驰 杜秀娟



摘要:针对随机背景噪声下的基音周期检测,提出了一种基于正交变换的基音检测算法。该算法通过正交变换对含噪的语音进行消噪,再求消噪后语音的自相关函数(ACF)和平均幅度差函数(AMDF),对其求出比值(ACF/AMDF)的平方以突显基音周期的峰值,以便获取较准确的基音周期参数。实验结果表明,与单一的自相关法和平均幅度差函数法相比,该算法获取的基音周期有较高的准确性和对噪声有较强的鲁棒性。
关键词:语音;正交变换;自相关函数;平均幅度差函数;基音周期检测
中图分类号:TN911 文献标识码:A 文章编号:1009-3044(2015)36-0126-03
Abstract: Aiming at pitch period detection under random background noise,a kind of pitch detection algorithm based on orthogonal transform is put forward. Noisy speech signal was first preprocessed by orthogonal transform, the autocorrelation function (ACF) and the average magnitude difference function (AMDF) for speech signal were obtained. The square of the ratio for ACF and AMDF had been applied to emphasize the peak of the true pitch period. This algorithm can get the exact pitch of the speech signal in strong noisy environment. Experimental results indicate that, compared with single autocorrelation and average magnitude difference method, the proposed algorithm has better robustness and higher accuracy.
Key words: speech; orthogonal transform; autocorrelation function; average magnitude difference function; pitch detection
1 概述
基音周期是语音信号处理及语音识别中的重要参数之一,基音周期参数的准确度越高便对语音编码、合成、识别等方面越有益。经典的基音周期提取算法有短时自相关法(ACF)、短时平均幅度差法(AMDF)、倒谱法[1-2]及其改进算法[3-5]。理论上讲,这些算法可以准确地提取纯净语音的基音周期。但语音通常会受到外界随机背景噪声的干扰,所以噪声环境下的基音周期检测是十分有必要的。因此,为了在随机背景噪声环境下提取较为准确的基音周期参数,研究人员们提出了很多新的基音周期检测算法,一种无门限U_V判决和基音检测算法[6],基于自相关平方函数与小波变换的基音检测[7],基于线性预测的综合基音检测法[8]以及利用小波变换加权自相关的基音检测法[9]等,这些算法在不同的应用范围内都取得了显著的成效。本文提出了一种基于正交变换和自相关函数、平均幅度差函数相结合的基音检测算法。……