一种语音特征参数子分量分析与有效性评价的新方法

语音信号中包含语义和说话人个性两大特征,其有效提取和强化对语音识别和说话人识别有着非常重要的意义。本文提出了一种语音特征参数中语义和个性特征子分量分析与有效性评价的4S方法,对语义和个性特征的成份比例进行分析,并通过量化指标评判特征参数对语音识别和说话人识别的有效性。运用4S分析方法对目前常用的特征参数LPC, LPCC和MFCC的子分量分析与有效性评价结果表明,所有的特征参数都更多地包含了语义特征信息,语义特征和说话人个性特征的成份比例因子LIR分别为1．30、1．44和1．61,并且,三种参数对语音识别和说话人识别的有效性均呈现出依次提高的特性。

A New Approach for Analyzing Subcomponents and Evaluating Effectiveness of Speech Feature Parameters

Linguistic message and speaker individuality are two main information in speech signals,how to detect or enhance these information subeomponents in speech feature parameters is very significant and challenge for speech and speaker reeoguition.In this paper,a novel approach named as 4S is proposed for analyzing the linguistics and individuality subcomponents in speech feature pa- rameters,and evaluating the effectiveness of feature parameters for speech and speaker recognition.The experiments on analysis and ef- fectiveness evaluation of usual feature parameters such as LPC,LPCC and MFCC based on 4S method show that,in these feature param- eters,the linguistic features are more contained than individualities with the linguistics individuality ratio factor LIR equaling to 1.30, 1.40,1.61 respectively,and the effectiveness of three parameters both for speech and speaker recognition are gradually higher.