多项式函数拟合实现汉语声调的语音合成

Speech synthesis of Chinese tones by polynomial function fitting

汉语语音的声调是个人语气与情感状态最直接的表达,是体现汉语语言状态最重要的特征之一。为了使得语音合成的逼真度得到有效的提高,讲话者的话语更加具有区分度,通过加入声调参数特征实现语音语调变换,以期成为情感识别和语音识别的准确度最有力的助推剂,弥补语音合成结果在情感特征以及语音演唱方面的不足。分别对汉语阴平、阳平、上声、去声采用基频提取的方式进行声调的分析、研究,将得到的基频曲线采用多项式函数拟合的方法对汉语4种声调进行重新构建,从数学角度对汉语声调进行分析、重构,采用三角函数曲线模拟随时间变化的语音基频曲线,根据共振峰频率将曲线进行叠加,达到了95.91%的满意的识别结果。结果表明:采用多项式函数拟合方法实现汉语4种声调的语音合成,更好地还原了语音的数学本质,使得抽象化的语音表现得更直观可控。

The tone of Chinese speech is the most direct expression of personal mood and emotional state,and it is one of the most important characteristics of the state of Chinese language.In order to effectively improve the fidelity of speech synthesis and make the speaker’s speech more distinguishable,the tone transformation is realized by adding tone parameter features,so as to become the most powerful booster for the accuracy of emotion recognition and speech recognition,and to make up for the shortcomings of speech synthesis results in emotion features and voice singing.The high and level tone,rising tone,falling-rising tone and falling tone of Standard Chinese are analyzed and examined by the way of fundamental frequency extraction.Finally,the method of polynomial function fitting is used to reconstruct the four tones of Chinese.The four tones are analyzed and reconstructed mathematically.Trigonometric function curve is used to simulate the fundamental frequency curves of Chinese tones with time.According to the formant frequency,the curves are superposed with the recognition result 95.91%.The synthesis results show that:the polynomial function fitting method may be used to realize speech synthesis of four tones in Chinese,which can better restore the mathematical nature of the voice and make the abstract speech more intuitive and controllable.