基于深度学习的中文情感语音合成方法  

Chinese emotional speech synthesis method based on deep learning

作　　者：王智 刘银华 WANG Zhi;LIU Yinhua(Institute of Future,School of Automation,Qingdao University,Qingdao,Shandong 266071,China)

机构地区：[1]青岛大学自动化学院未来研究院,山东青岛266071

出　　处：《自动化与仪器仪表》2022年第9期10-15,共6页Automation & Instrumentation

摘　　要：语音合成是人机交互的组成部分,并在该过程中起到了闭环的作用。人在说话时能表现出自己的情绪状态,如高兴、悲伤、愤怒等,在现有的语音合成模型中没有得到充分体现,致力于合成出具有情感特征的中文语音,提出了一种中文情感语音合成方法,该方法结合情感语音库对模型优化训练,添加中文处理模型,通过对语音参数的修正,可以提高语音的情感度。结果表明:成熟的模型可以合成出优质的中文语音,情感方面也得到有效的体现。Speech synthesis is an integral part of human-computer interaction and plays a role in closing the loop in the process. People can show their emotional states when speaking, such as happy, sad, angry, etc., which are not fully reflected in existing speech synthesis models. Dedicated to synthesizing Chinese speech with emotional features, we propose a Chinese emotional speech synthesis method, which combines an emotional speech library to optimize the training of the model, add Chinese processing model, and through the correction of speech parameters, can improve the emotionality. The results show that the mature model can synthesize high quality Chinese speech and the emotional aspect is effectively reflected.

关 键 词：语音合成 情感 深度学习 神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]