论文部分内容阅读
在实际语音情感识别系统中,训练语音和测试语音往往来自不同的语料库,识别率下降显著。针对这一问题,该文提出一种有效的基于特征迁移学习的跨库语音情感识别方法。引入最大均值差异(maximum mean discrepancy,MMD)来描述不同数据库情感特征分布之间的相似度,并通过最大均值差异嵌入(maximum mean discrepancy embedding,MMDE)算法及特征降维算法来寻找二者之间的邻近低维特征空间,并在此低维空间中训练得到情感分类器用于情感识别。同时为了更好地保证情感信息的类别区分度,进一步引入半监督判别分析(semi-supervised discriminant analysis,SDA)方法用于特征降维。最后在2个经典语音情感数据库上对提出的方法进行实验评价,实验结果表明:提出的方法可以有效提高跨库条件下的语音情感识别率。
In the actual speech emotion recognition system, training speech and test speech often come from different corpora, the recognition rate drops significantly. To solve this problem, this paper proposes an effective cross-database speech recognition method based on feature-based learning. The maximum mean discrepancy (MMD) is introduced to describe the similarity between the emotional feature distributions of different databases. The maximum mean discrepancy embedding (MMDE) algorithm and feature dimensionality reduction algorithm are used to find the similarity between the two. , And the emotion classifier is trained in this low-dimensional space for emotion recognition. At the same time, in order to ensure the classification of affective information better, a semi-supervised discriminant analysis (SDA) method is further introduced for feature dimensionality reduction. Finally, the proposed method is evaluated experimentally on two classic speech emotion databases. The experimental results show that the proposed method can effectively improve the recognition rate of speech emotion under cross-library conditions.