一种基于小波变换的清浊音分类基音检测方法  

A Method of Unvoiced/Voiced Classification and Pitch Detection Based on Wavelet Transform

作　　者：胡瑛[1] 陈宁[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《电声技术》2006年第11期63-66,共4页Audio Engineering

摘　　要：提出了一种基于小波变换的鲁棒性基音周期检测方法。首先结合平均能量频带分布和短时过零率这两个特征参数对语音信号进行清浊音判决,然后对浊音段采用空域相关函数提取基音周期。实验表明,与传统的小波变换和自相关算法相比,该方法鲁棒性好,对基音检测具有更高的准确性。A robust pitch detection method based on wavelet transform is proposed. Firstly, a new algorithm for the classification of speech into voiced and unvoiced segments based on statistical analysis of the frequency distribution of the average energy in the wavelet domain and the short-time zero-crossing rate of the speech signal is proposed. Then an algorithm based on spatial correlation function for estimating pitch frequency only in voiced regions is presented. The experimental results show that this algorithm has a better robustness and increases the precision of pitch detection compared with the classical wavelet-based methods and auto correletion function.

关 键 词：平均能量频带分布 短时过零率 小波变换 空域相关函数 

分 类 号：TN912[电子电信—通信与信息系统]