基于重要性评分的多级随机森林网络语音情感识别

Multi-level random forest speech emotional recognition based on importance score

在源数据不充分或不平衡的情况下,深度学习方法在小样本集上难以取得令人满意的语音情感识别效果。因此,本研究构造了一种三层随机森林情感识别网络,在每一层都单独剥离易于区分的情感类别,并通过重要性评分方法,为每一层网络都构造一个识别特定类别的特征集,该特征集的每一个特征都依据贡献度大小得到赋权,以确保对分类贡献越多的特征因子对结果影响越大。本研究构建的多级情感识别网络,在小样本集语音情感识别的整体识别率上,较单层随机森林网络和支持向量机分别提高了5%和7%,较流行的深度学习方法卷积神经网络提高了12%。实验结果和理论分析表明:基于重要性评分的多级随机森林网络相较于其他方法,在源数据样本量较少和部分不平衡的情况下,有更高的识别准确率,具有语音情感识别方向的实际应用意义。

In the case of insufficient or unbalanced source data, deep learning method is difficult to achieve satisfactory emotion recognition effect on small sample set. Therefore, this paper constructs a three-layer random forest emotion recognition network, which separates the easily distinguishable emotion categories in each layer, through the importance scoring method. A feature set for identifying specific categories is constructed for each layer of the network, each feature of the feature set is weighted according to the contribution degree to ensure that the feature factor pairs that contribute more to the classification result, the greater the impact. The multi-level emotion recognition network constructed in this paper improves the overall recognition rate of speech emotion recognition in small sample sets by 5% and 7% respectively compared with single-layer random forest network and support vector machine, and the network increased by 12% compared to popular deep learning method CNN. The experimental results and theoretical analysis show that the multi-level random forest network based on importance score has higher recognition accuracy and speech emotion recognition than other methods when the source data sample size is small and partially unbalanced, so that it has the practical significance of the direction.