函数拟合实现带声调的语音合成  

SPEECH SYNTHESIS WITH TONE BY FUNCTION FITTING

作　　者：李建文[1] 王咿卜 Li Jianwen;Wang Yibo(School of Electronic Information&Artificial Intelligence,Shaanxi University of Science&Technology,Xi an 710021,Shaanxi,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院,陕西西安710021

出　　处：《计算机应用与软件》2022年第9期193-200,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(60672001)。

摘　　要：语音的声调最能突显个人语气与情感状态,通过加入声调参数这一特征,能够有效提高语音合成的逼真度,使得话语更加具有区分度,并且有效提高了情感识别和语音识别的准确度,弥补了语音合成的结果缺乏情感特征以及语音演唱的不足。采用基频提取的方式分别对汉语一声、二声、三声、四声声调进行分析、研究,最终采用多项式函数拟合的方法进行声调的重新构建,从数学角度对语音声调进行分析、重构,采用三角函数模拟不同时间的语音基频曲线,将不同频率的曲线进行叠加,达到了95.91%的满意识别结果。合成结果表明:采用函数拟合方法进行带声调的语音合成,更好地还原了语音的数理本质,使得抽象化的语音表现得更为具体、可控。The tone of voice can highlight personal mood and emotional state.By adding the feature of tone parameter,it can effectively improve the fidelity of speech synthesis,make speech more distinguishable,improve the accuracy of emotion recognition and speech recognition,and make up for the lack of emotion feature and voice singing in the result of speech synthesis.The first tone,second tone,third tone and fourth tone of Chinese were analyzed and studied by the way of fundamental frequency extraction,and the method of polynomial function fitting was used to reconstruct the tone.From the mathematical point of view,the tone was analyzed and reconstructed.Trigonometric function was used to simulate the fundamental frequency curves of different times,and the curves of different frequencies were superposed.It achieved 95.91%satisfactory recognition results.The synthesis results show that the function fitting method for tone speech synthesis can better restore the mathematical essence of speech and make the abstract voice performance more concrete and controllable.

关 键 词：软件工程 语音合成 函数拟合 基频提取 声调 情感 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术] TN912.33[电子电信—通信与信息系统]