【作者】 崔宣；

【导师】 潘世永；

【作者基本信息】 西华大学 ， 计算机软件与理论， 2008， 硕士

【摘要】 说话人识别是指通过说话人语音信号的分析和特征提取,从而确定说话人是否在所记录的说话人集合中,进而确定说话人是谁的过程。它在许多领域内有良好的应用前景。目前在说话人识别中,要提高识别率有两个重要的问题需要解决:一是如何选取能够有效表征说话人特征的可靠参数;二是如何选取合适的识别算法。本文主要是对特征参数的选取进行了初步的探讨,做了如下几方面工作:1.在特征提取方面,本文中分析了当前最常用的两种倒谱特征参数:美尔频率倒谱系数(MFCC)和线性预测倒谱系数(LPCC)。并对其进行了改进,一方面是采用二次提取的方法,将MFCC和LPCC与其各自对应的一阶差分组合在一起形成新的特征参数。另一方面是本文还提出了将美尔频率倒谱系数(MFCC)和线性预测倒谱系数(LPCC)两个基于不同模型的特征参数组合在一起形成新的特征参数,实验的结果证明了这两种方法与传统的使用单一特征参数进行识别相比都能有效的提高实验系统的识别率。此外,还尝试着在预处理部分加入基于时域特征的端点检测,使用到了短时能量参数和短时过零率相组合,然后在特征提取部分,提取20阶MFCC作为特征参数来进行识别,但实验的结果没有达到理想效果。2.在识别算法方面,本文对矢量量化的方法进行了研究,并用matlab语言实现了一个有效的说话人辨认识别系统。更多还原

【Abstract】 The speaker recognition is the processing of automatically recognition whether the speaker in the speakers which records gathers, then determined who the speaker is, by analyzing the speaker’s pronunciation signals and picking up the speaker’s characteristic. Now, it has well application prospects in many fields.Currently in the field of speaker recognition, there are two important questions need to solve for the enhancement of recognition rate. For one hand is how to select more effective and reliable speaker characteristic, For anther hand is how to select the best recognition methods. This article mainly has made the discussion in the fist question, and made following aspects improvement and research.Firstly, we propose that choose MFCC ,LPCC and MFCC and LPCC’s difference to be the new speech characteristic parameters .Using VQ to recognize text-inpendent speech ,we have developed a speaker identification in this paper ; We do experiment, mixing MFCC and LPCC together ,to make a new characteristic parameter ,it is prove that ,it can effectively improve the rate of the system recognition ; We also make a test with the extreme point detection before pretreatment and we choose the MFCC as characteristic ,but it’s a pity the result is failure.Secondly, do some research of the VQ method, and apply the speakeridentification experiment in this paper更多还原