论文部分内容阅读
针对情感语音合成系统中情感的细腻刻画与自动预测问题,提出多视角情感描述模型,从认知评价、心理感受、生理反应和发音方式4个方面刻画言语情感的产生过程和衍化机制;引入能够支持分布式特征且具有堆叠结构的多层神经网络——深层堆叠网络构建从文本到情感描述的预测模型。实验结果表明在预测模型中引入不同情感成分和上下文信息作为特征有助于提升预测效果,验证了采用深层堆叠网络进行情感预测的有效性与多视角情感描述模型的合理性。
Aiming at the delicate characterization and automatic prediction of emotion in speech synthesis system, a multi-perspective emotion description model is proposed to describe the generation process and derivation mechanism of speech emotion from four aspects: cognitive evaluation, psychological feeling, physiological reaction and pronunciation way. Multilayer Neural Networks with Distributed Features and Stacked Structures - A Deep Stacked Network Constructs a Predictive Model from Text to Sentiment Description. The experimental results show that introducing different affective components and contextual information into the prediction model can improve the prediction performance and verify the validity of using deep-layer network for the prediction of emotion and the multi-view emotional description model.