摘要： 为提高语音人格的预测精度,结合随机森林模型,提出一种语音人格预测方法。选取用于语音人格预测的候选韵律特征集,通过Bootstrap方式对语音韵律特征集进行抽样,根据基尼系数为每个决策树节点选择最优韵律特征集,最终构造各维人格特征对应的随机森林模型,实现语音人格预测。在公共语音人格预测数据集中的仿真实验结果表明,与其他语音人格预测方法相比,该方法具有更高的预测准确率。

关键词: 语音人格, 人格预测, 大五人格理论, 随机森林, 韵律特征

Abstract: In order to improve the prediction accuracy of speech personality,this paper proposes a method of speech personality prediction based on random forest model.The candidate prosodic feature set for speech personality prediction is selected,and the prosodic feature set is sampled by bootstrap method.According to the Gini coefficient,the optimal prosodic feature set is selected for each decision tree node,and the random forest model corresponding to each dimension personality is constructed to achieve speech personality prediction.The results of experiment on open source speech personality corpus show that this method has better performance in terms of prediction accuracy than other speech personality prediction methods.

Key words: speech personality, personality prediction, big five personality theory, random forest, prosodic feature

中图分类号: 

• TP37