采用原子表示模型的维吾尔语语音情感识别

针对现有的基于表示学习的语音情感计算算法中存在着限制条件单一的问题,且没有证明它们的有效性,提出了一种采用原子表示模型的语音情感识别算法。通过引入一个新的条件,称为原子分类条件。在这种条件下,对正确识别新的测试情感样本有较好的效果。现有的基于表示的分类算法以单一的稀疏表示方法为主,而提出的算法可以结合稀疏表示模型和其他的表示模型。该算法能够放宽适用条件的范围,使得原子表示模型适应更多分类任务。采集并建立了维吾尔语语音情感数据库。在该情感数据库上,分析维吾尔语情感语音的基本声学特征。通过对情感特征空间进行原子表示的映射变换,可以有效表示情感特征空间。经实验结果证明所提出的方法优于传统的方法,在维吾尔语情感语音数据库上达到了64.17%识别率。 

Atomic Representation based Emotion Recognition from Uyghur Speech

Aiming at the problem of single restrictive condition in existing speech emotion computing algorithms based on representation learning, and that few results are reported to justify their effectiveness. A speech emotion recognition algorithm based on atomic representation model is proposed. By introducing a new condition, called atomic classification condition. In such a condition, recognition capacity to new emotion samples was improved. The existing representation based classification algorithms are mainly sparse representation methods. The proposed algorithm can combine sparse representation model with other representation models. The algorithm can relax the scope of application conditions and adapt the representation model to more classification tasks. An Uyghur emotional speech database was established.Based on the Uyghur speech emotion database, the basic acoustic characteristics of Uyghur emotional speech are analyzed. Emotional feature space can be effectively represented by mapping of atomic representation. The experimental results show that the proposed method is better than the traditional method, and the recognition rate reached 64.17% on the Uyghur emotional speech database.