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。
A multi-perspective emotion model is presented to provide more details about the emotions in expressive speech synthesis and to facilitate automatic predictions. The method describes the emotion development in terms of the cognitive appraisal, psychological feeling, physical response and utterance manner. The descriptive model is used to develop a text-based emotion prediction model using a deep neural network (the deep stacking network), which supports distributed representation and has a stacking structure. Tests validate the benefits of using this prediction method for the interactions among different emotional aspects and the contextual impacts, as well as the effectiveness of the deep stacking network and the multi-perspective emotion model.
[1] Govind D, Prasanna S R M. Expressive speech synthesis:A review[J]. International Journal of Speech Technology, 2013, 16(2):237-260. [2] 徐俊, 蔡莲红. 面向情感转换的层次化韵律分析与建模[J]. 清华大学学报:自然科学版, 2009, 49(S1):1274-1277.XU Jun, CAI Lianhong. Hierarchical prosody analysis and modeling for emotional conversions[J]. J Tsinghua Univ:Sci & Tech, 2009, 49(S1):1274-1277. (in Chinese) [3] TAO Jianhua, KANG Yongguo, LI Aijun. Prosody conversion from neutral speech to emotional speech[J]. IEEE Transactions on Audio, Speech, and Language Processing, 2006, 14(4):1145-1154. [4] 韩纪庆, 邵艳秋. 基于语音信号的情感处理研究进展[J]. 电声技术, 2006(5):58-62.HAN Jiqing, SHAO Yanqiu. Research progress of emotion processing based on speech signal[J]. Audio Engineering, 2006(5):58-62. (in Chinese) [5] Ekman P, Friesen W V, O'Sullivan M, et al. Universals and cultural differences in the judgments of facial s of emotion[J]. Journal of Personality and Social Psychology, 1987, 53(4):712-717. [6] Cowie R, Douglas-Cowie E, Savvidou S, et al. FEELTRACE:An instrument for recording perceived emotion in real time[C]//ISCA Tutorial and Research Workshop (ITRW) on Speech and Emotion. Newcastle, UK, 2000:19-24. [7] Mehrabian A. Pleasure-arousal-dominance:A general framework for describing and measuring individual differences in temperament[J]. Current Psychology, 1996, 14(4):261-292. [8] Moors A, Ellsworth P C, Scherer K R, et al. Appraisal theories of emotion:State of the art and future development[J]. Emotion Review, 2013, 5(2):119-124. [9] 高莹莹, 朱维彬. 言语情感描述体系的试验性研究[J]. 中国语音学报, 2013, 4:71-81.GAO Yingying, ZHU Weibin. The research for the description system of speech emotion[J]. Chinese Journal of Phonetics, 2013, 4:71-81. (in Chinese) [10] DENG Li, YU Dong, Platt J. Scalable stacking and learning for building deep architectures[C]//IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Kyoto, Japan, 2012:2133-2136. [11] Riedl M, Biemann C. Text segmentation with topic models[J]. Journal for Language Technology and Computational Linguistics, 2012, 27(1):47-69. [12] Blei D M, Ng A Y, Jordan M I. Latent Dirichlet allocation[J]. The Journal of Machine Learning Research, 2003, 3:993-1022. [13] Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786):504-507. [14] Hinton G. A practical guide to training restricted Boltzmann machines[J]. Momentum, 2010, 9(1):599-619. [15] YU Dong, DENG Li. Accelerated parallelizable neural network learning algorithm for speech recognition[C]//12th Annual Conference of the International Speech Communication Association (INTERSPEECH). Florence, Italy:ISCA Press, 2011:2281-2284.
