基于MCE准则的语音识别特征线性判别分析

提出了一种基于最小分类错误（Minimum classification error，MCE）准则的线性判别分析方法（Linear discriminant analysis，LDA），并将其应用到连续语音识别中的特征变换.该方法采用非参数核密度估计方法进行数据概率分布估计；根据得到的概率分布，在最小分类错误准则下，采用基于梯度下降的线性搜索算法求解判别分析变换矩阵.利用判别分析变换矩阵对相邻帧梅尔滤波器组输出拼接的超矢量变换降维，得到时频特征.实验结果表明，与传统的MFCC特征相比，经过本文判别分析提取的时频特征其识别准确率提高了1.41%，相比于HLDA（Heteroscedastic LDA）和近似成对经验正确率准则（Approximate pairwise empirical accuracy criterion，aPEAC）判别分析方法，识别准确率分别提高了1.14%和0.83%.

A Minimum Classification Error Criterion Based Linear Discriminant Analysis Method for Speech Recognition Feature

A linear discriminant analysis (LDA) method based on the minimum classification error criterion is proposed, and further it is applied to the continuous speech recognition feature transformation. The data probability distribution is estimated using non-parametric kernel density estimation method. According to the obtained probability distribution, a gradient descent based linear search procedure is performed to get the discriminant analysis transformation matrix under the minimum classification error criterion. The dimensionality of super-vector conjoined by the adjacent frames Mel filter bank output is reduced with the transformation matrix, and then after dimensionality reduction the time-frequency feature is acquired. Experimental results show that compared with the traditional MFCC feature, the recognition accuracy rate of the time-frequency feature extracted with the presented discriminant analysis method has a 1.41% improvement. In contrast with the HLDA and aPEAC discriminant analysis feature transformation method, the recognition accuracy of the presented method increases by 1.14% and 0.83% separately.